CN106408017A - Ultrasonic carotid artery intima media thickness measurement device and method based on deep learning - Google Patents

Abstract

The invention relates to medical image processing, in order to provide an IMT measurement algorithm with better overall performance, that is, without human participation, relatively low time complexity, high consistency between automatic measurement results and expert manual measurement results, and for different morphological structures Satisfactory measurement results can be obtained for the intima. The method for measuring the intima-media thickness of the ultrasonic carotid artery based on deep learning in the present invention has the following steps: 1) ROI acquisition: the convolutional neural network (CNN) is used to automatically identify the distant part of the ultrasonic carotid artery. 2) Pixel classification: perform more advanced feature extraction, and add a logistic regression classification layer after feature extraction to build a pattern classifier to complete the classification of image pixels; 3) Boundary extraction: use the area information and position of the target classification area The classification area is selected, and the polynomial curve fitting is performed on the classification boundary. The invention is mainly applied to medical image processing.

Description

Ultrasonic carotid artery intima-media thickness measurement device and method based on deep learning

technical field

The present invention relates to medical image processing, in particular, to a device and method for measuring the intima-media thickness of an ultrasonic carotid artery based on deep learning.

Background technique

Cardiovascular and cerebrovascular diseases have become the primary diseases that endanger human health. According to the survey of the World Health Organization, the number of people who died of cardiovascular and cerebrovascular diseases in the world in 2012 was as high as 17.5 million, accounting for about 30% of the total global death population. Atherosclerosis is the primary cause of cardiovascular and cerebrovascular diseases. It is mainly manifested by the deposition of lipids and cholesterol in the blood on the inner wall of blood vessels, which makes the intima media thicker and less elastic. This pathological condition may last for many years and It is not discovered, but plaque in blood vessels can lead to complete blockage of blood vessels, resulting in stroke and myocardial infarction. Therefore, early follow-up diagnosis of atherosclerosis is extremely important for the prevention of cardiovascular and cerebrovascular diseases. Numerous studies have shown that carotid artery intima-media thickness (Intima MediaThickness, IMT), as shown in Figure 1, lumen-intima boundary (Lumn-Intima Interface, LII) and media-adventitia boundary (Media-Adventitia Interface, MAI ) can effectively reflect the degree of atherosclerosis and is an important indicator for predicting cardiovascular and cerebrovascular diseases.

Ultrasound images are widely used in clinical practice because of their non-invasive, low price, and fast imaging. In order to ensure the reproducibility of IMT measurement, IMT was measured at the distal end of the carotid artery without plaque, generally at least 5 mm below the carotid bifurcation, where there was an obvious bilinear structure.

Clinically, the intima-media thickness was calculated manually by medical staff, and by drawing the boundaries of LII and MAI. Due to the different levels of training and experience of operators, the definition of boundaries is affected by subjective cognition, resulting in different measurement results obtained by different operators, that is, there are differences between observers; even if the same operator, the results measured at different times There will also be differences, that is, intra-observer errors, which are not repeatable, time-consuming and cumbersome, and cannot meet the needs of massive data processing in modern medicine. Based on the above reasons, it is necessary to implement a fully automatic, fast, accurate and robust IMT measurement algorithm using digital image processing technology.

In 1986, the first IMT segmentation algorithm was proposed, and a large number of algorithms and schemes were proposed successively in the following 30 years. The IMT measurement scheme is roughly composed of two parts. The first step is to extract the region of interest (Region of Interest, ROI) and obtain the initial contour line, and the second part is to obtain the precise contour line. As far as the first step is concerned, according to the degree of manual intervention, these existing algorithms can be divided into fully automatic measurement algorithms and semi-automatic measurement algorithms. Semi-automatic segmentation requires artificial selection of ROI or points of interest, while fully automatic methods do not require human intervention. Some scholars use level sets to obtain initial contours and obtain ROIs. The methods of watershed and template matching are also applied to the acquisition of ROIs; The second step involves dynamic programming, active contour models, neural network models, statistical modeling and many other theories.

Contents of the invention

In order to overcome the deficiencies in the prior art, the present invention aims to provide an IMT measurement algorithm with better overall performance, that is, no human participation is required, the time complexity is relatively low, and the automatic measurement results are consistent with the expert manual measurement results. Satisfactory measurement results can be obtained for the inner seed membranes with different morphological structures. For the first stage, the present invention applies the convolutional neural network to automatically identify the distal end of the supercarotid artery, and then extracts the ROI. Then, the pattern classifier constructed by stacked autoencoder and logistic regression layer is used to classify the pixels, and the area information and position information of the classification area are used to screen the classification results, remove the wrongly classified pixels, and select a reliable classification area. Using curve fitting to extract the boundary to complete the IMT measurement. The technical scheme adopted in the present invention is a device and method for measuring intima-media thickness of the ultrasonic carotid artery based on deep learning, and the steps are as follows:

1) ROI acquisition: the convolutional neural network CNN (Convolution Neural Network) is used to automatically identify the distal end of the ultrasonic carotid artery, and then extract the ROI;

2) Pixel classification: The pixel gray value information of a certain area centered on the pixel to be classified is used as the input of the stacked autoencoder, and its output is used as the input for higher-level feature extraction. After the feature extraction, a logistic regression classification layer is added to construct a pattern classifier , to complete the classification of image pixels;

3) Boundary extraction: Use the area information and position of the target classification area to select the classification area. Use the row index value of the farthest end of the carotid artery and the barycenter row value and area information of the area to be determined to remove misclassified areas that are close to the target area but have a large area, and use the center of gravity between the largest area of the target and the remaining smaller areas to be determined Compare the row values, combine the area information of the smaller area to remove the area that is farther away from the target area and have a smaller area, and finally index by column to remove the area with incomplete intima-media structure; Perform a polynomial curve fit.

The specific steps of ROI extraction using convolutional neural network CNN are as follows:

(1) Crop the image, and cut out the information irrelevant to the image analysis in the image;

(2) the cropped image is evenly divided into five equal columns, and image blocks of a certain size are sequentially taken out along the symmetrical axis of the equally divided sub-image;

(3) The image block is used as the input of the trained CNN for predictive classification, the number of categories is 2, that is, the image containing the "dark-bright-dark-bright" structure and the image without this structure, and the same sub-image is selected The middle attribute contains the image block with the largest predicted value of the "dark-bright-dark-bright" structure class;

(4) Sorting the row index values of the obtained image blocks, selecting the row index values that can effectively calibrate the distal end of the carotid artery, and extracting the ROI based on the index values, the selection rules are as follows.

r1, r2 and r3 are the maximum value, the second maximum value and the middle value of the row index value respectively. m1, m2 are the mean values of r1, r2 and r3, r2 respectively, v is a set threshold, and r3 is preferred as the effective index value of the distal end of the carotid artery. If r3 is less than a certain threshold, it is considered that r3 cannot effectively define the distal end of the carotid artery. At this time, use m1 or m2.

The algorithm execution process of the Auto Encoder AE (Auto Encoder) includes an encoding process and a decoding process. Its operation process is shown in formula (1) (2):

y＝f(w _y *x+b _y ) (1)

z＝f(w _z *y+b _z ) (2)

In the formula, x represents the input of AE, and also represents the output of the hidden layer, z represents the output of AE, w _y , b _y are the weights and biases from the input layer to the hidden layer, w _z , b _z are the weights and biases from the hidden layer to the hidden layer, respectively. The weight and bias of the output layer, f(.) is the activation function;

The training process of AE is the process of minimizing the error function,

In order to reduce the number of training parameters, usually set

W _y =W _z =W (3)

Choose the following rules to update the weight term and bias term to minimize the error function:

The stacked autoencoder SAE (Stacked Auto Encoder) is composed of a multi-layer autoencoder, the output of the previous layer of autoencoder is used as the input of the next layer of autoencoder, and the activation value vector of the deepest hidden layer unit is the input higher-order representation of values;

Select the sigmoid function as the activation function of the codec. In the pre-training process, the network mainly learns three parameters W, b _y , b _z . Before setting the parameter update rules, it is necessary to reasonably determine the error term, that is, the cost function. Select Cross entropy is used as a cost function, and at the same time the entire training data is divided into small pieces of data, so for the entire data set, the parameters are updated with small pieces of data:

In the formula, d is the dimension of the input vector, m is the size of the data block, x _ik and z _ik are the kth component of the i-th input data in the selected block respectively;

To optimize formula (8) with stochastic gradient descent method, first express the reconstruction layer in the form of scalar:

Among them, net _ip ^y is the p-th dimension of the hidden layer unit, netikz i is the p-th dimension of the output unit, which represents the number of the block data, d represents the dimension of the input data, h represents the dimension of the hidden layer, z _ik is the dimension of the input data k-th dimension reconstruction;

The first and second derivatives of (7) are:

f'(x)=f(x)1-f(x) (12)

f"(x)=f(x)1-f(x)1-2f(x) (13)

According to (9)-(11), calculate the partial derivative of the reconstruction layer with respect to w, b _y , b _z ;

In the formula, Wrs is the weight connecting the rth input and the sth hidden layer unit, b _yr is the bias of the rth unit in the hidden layer, and b _zr is the bias of the rth unit in the reconstruction layer;

The partial derivatives of the cost function with respect to W, b _y , b _z are obtained from equations (9)-(16):

After the network training is completed, the data reconstruction of the network is removed, the output of the hidden layer is the learned feature, and the output of the subsequent network layer is the input of the previous layer, and it is trained in the same way, that is, the network is greedily trained layer by layer, so as to avoid the loss of the network due to the initial weight loss. The value is too small to fall into the local optimal solution; finally, combine the layers of the network together, use the features learned by the network to classify, and add a soft-max classification layer after the pre-training network to fine-tune the entire pre-training network to complete the classification task. Given a set of input data, the probability that the input belongs to a certain class i is equal to

In the formula, R is the input of the logistic regression layer, W, b are the weight and bias of the logistic regression layer, and the sum of all outputs is 1.

The logistic regression layer and the stacked autoencoder form a deep classifier. The process of training the deep classifier is as follows:

Build two different pattern classifiers according to actual needs, denoted as SAE_NB and SAE_LM, respectively use the above training rules to train two deep pattern classifiers in the selected representative training set, and use SAE_NB to divide pixels into boundaries pixels and non-boundary pixels, and then use SAE_LM to divide the boundary pixels into LII and MAI pixels.

According to the principle of the minimum sum of squared deviations, polynomial curve fitting is performed on the classification boundary, and the process is as follows:

Let the fitting polynomial be:

f＝a ₀ +a ₁ e+...+a _k e ^k (21)

Among them, e and f are the column value and row value of the boundary point pixel respectively, a _k is the polynomial coefficient, and k is the polynomial degree.

(2) The sum of the distances from each point to this curve, that is, the square of the deviation is as follows:

Among them, n is the size of the data set, (e _i , f _i ) is the column and row values of the i-th pixel, and R represents the deviation.

(3) Calculate the partial derivative of a _i on the right side of the equation to get

(4) Simplify the left formula to get the following formula

(5) Express the equation in matrix form

(6) Simplify the Vandermonde matrix obtained in (5) to get

(7) (6) formula is E*A=F, then

A＝(E*E) ^-1 *E*F (27)

When the coefficient matrix is obtained, the fitting curve is obtained at the same time.

Features and beneficial effects of the present invention are:

The invention can effectively solve the problem of generally poor overall performance in current IMT algorithms, and can realize fully automatic, fast, accurate and robust measurement of IMT.

In the ROI selection stage, the CNN is used to identify the distal end of the ultrasonic carotid artery. CNN has a certain tolerance to the input data with large deformation and has certain robustness to noise; Different images can accurately locate the distal end of the carotid artery and correctly extract the ROI.

In the pixel classification stage, according to the different characteristics of the classification set, different stacked autoencoders are built; the stacked autoencoder extracts the features of the input data layer by layer, and finally obtains the deep features of the data and classifies them to ensure the efficiency of classification. sex and effectiveness.

In the boundary extraction stage, considering the incomplete intima-media structure of the ultrasound images used, it is inevitable that there will be fractures. When selecting reliable classification regions, not only the area of the region but also the location of the region should be considered.

Description of drawings:

Figure 1 Ultrasound axial section of the carotid artery.

Figure 2. Schematic diagram of a single-layer autoencoder.

Fig. 3 is an algorithm flow chart adopted by the present invention.

Figure 4. Schematic diagram of pattern classifier classifying pixels.

detailed description

Although IMT segmentation algorithms have made breakthroughs in recent years, it is difficult for these existing algorithms to meet the requirements of full automation, speed, accuracy, and robustness at the same time. The present invention aims to develop an IMT measurement algorithm with better overall performance, that is, without human participation, relatively low time complexity, high consistency between automatic measurement results and expert manual measurement results, and uniform for internal seed films with different morphological structures. Satisfactory measurement results can be obtained. For the first stage, the present invention applies the convolutional neural network to automatically identify the distal end of the superborn carotid artery, and then extracts the ROI. Then, the pattern classifier constructed by stacked autoencoder and logistic regression layer is used to classify the pixels, and the area information and position information of the classification area are used to screen the classification results, remove the wrongly classified pixels, and select a reliable classification area. Using curve fitting to extract the boundary to complete the IMT measurement.

1ROI acquisition

In order to ensure fully automatic, fast and accurate ROI acquisition, the present invention adopts a convolution neural network (ConvolutionNeural Network, CNN) to automatically identify the distal end of the ultrasonic carotid artery vessel, and then extract the ROI. The specific steps are as follows:

(1) Crop the image, cut out the information irrelevant to the image analysis in the image, such as the personal information of the subject contained in the edge of the original ultrasound image.

(2) Evenly divide the cropped image into five equal columns. Image blocks of a certain size are sequentially taken out along the symmetry axis of the equally divided sub-image.

(3) The image block is used as the input of the trained CNN for predictive classification, the number of categories is 2, that is, the image containing the "dark-bright-dark-bright" structure and the image without this structure, and the same sub-image is selected The medium belongs to the image block containing the largest predicted value of the "dark-bright-dark-bright" structure class.

(4) Sorting the row index values of the obtained image blocks, selecting the row index values that can effectively calibrate the distal end of the carotid artery, and extracting the ROI based on the index values, the selection rules are as follows.

r1, r2 and r3 are the maximum value, the second maximum value and the middle value of the row index value respectively. m1, m2 are the mean values of r1, r2 and r3, r2 respectively. v is a set threshold. In the present invention, r3 is preferred as the effective index value of the distal end of the carotid artery. Generally, the distal end of the carotid artery is located in the middle and lower position of the entire ultrasound image. Therefore, if r3 is less than a certain threshold, r3 is considered to be ineffective. Define the distal end of the carotid artery, this time with m1 or m2.

2 pixel classification

The deep learning architecture is composed of multi-layer nonlinear computing units. Studies have shown that the deep feature expression extracted by the function family is more effective than the shallow feature extracted by a single function. Stacked Autoencoder (SAE) is a deep neural network built with a numerical model, and its hidden layer nodes are computational units with practical significance. For numerical data with continuity and determinism, the numerical model will obtain better training results.

Auto Encoder (AE) is an unsupervised learning algorithm that uses the backpropagation algorithm and makes the target value equal to the input value, ie z≈x.

A schematic diagram of a single-layer autoencoder is shown in Figure 2.

The algorithm execution process of AE includes encoding process and decoding process. Its operation process is shown in formula (1) (2):

y＝f(w _y *x+b _y ) (1)

z＝f(w _z *y+b _z ) (2)

In the formula, wy and by are the weight and bias from the input layer to the hidden layer respectively, wz and bz are the weight and bias from the hidden layer to the output layer respectively, and f(.) is the activation function.

The training process of AE is the process of minimizing the error function,

In order to reduce the number of training parameters, usually set

W _y =W _z =W (3)

Generally, the following rules are selected to update the weight term and bias term to minimize the error function:

The stacked autoencoder neural network is a neural network composed of multi-layer autoencoders, the output of the previous layer of autoencoder is used as the input of the next layer of autoencoder, and the activation value vector of the deepest hidden unit is the input A higher-order representation of a value.

The invention uses the relevant technology of pattern recognition to solve the problem of image pixel classification and completes image segmentation. Specifically, the pixel gray value information of a certain area centered on the pixel to be classified is used as the input of the stacked autoencoder, and its output is used as the input for higher-level feature extraction. After the feature extraction, a logistic regression classification layer is added to construct a pattern classifier, and the model is completed. Classification of image pixels. The sigmoid function can perform nonlinear mapping, its output range is limited, and the data is not easy to diverge in the process of transmission, so the sigmoid function is selected as the activation function of the codec in this paper. During the pre-training process, the network mainly learns three parameters W, by, bz. Before setting the parameter update rules, it is necessary to reasonably determine the error term, that is, the cost function. Because the activation function uses the sigmoid function, the derivative of the cost function will approach 0 because the neuron output is 0 or 1. If cross-entropy is used, even if the network nodes tend to be saturated, the parameters will continue to be updated, and finally cross-entropy is selected as the cost function. At the same time, the entire training data is divided into small pieces of data, so for the entire data set, the parameters are updated with small pieces of data.

where d is the dimension of the input vector, m is the size of the data block, and x _ik (zi _ik ) is the kth component of the i-th input data in the selected block.

To optimize formula (8) with stochastic gradient descent method, first express the reconstruction layer in the form of scalar:

Among them, net _ip ^y (net _ik ^z ) is the p-th dimension of the hidden layer (output) unit, i represents the number of the block data, d represents the dimension of the input data, and h represents the dimension of the hidden layer. z _ik is the k-th dimensional reconstruction of the input data.

The first and second derivatives of (7) are:

f'(x)=f(x)1-f(x) (12)

f"(x)=f(x)1-f(x)1-2f(x) (13)

According to (9)-(11), calculate the partial derivative of the reconstruction layer with respect to w, b _y , b _z ;

In the formula, Wrs is the weight connecting the rth input and the sth hidden layer unit, b _yr (b _zr ) is the bias of the rth unit in the hidden layer (reconstruction layer).

The partial derivatives of the cost function with respect to W, b _y , b _z are obtained from equations (9)-(16):

After the network training is completed, the data reconstruction of the network is removed, the output of the hidden layer is the learned feature, and the output of the subsequent network layer is the input of the previous layer, and it is trained in the same way, that is, the network is greedily trained layer by layer, so as to avoid the loss of the network due to the initial weight loss. The value is too small to fall into the local optimal solution; finally, combine the layers of the network together, use the features learned by the network to classify, and add a soft-max classification layer after the pre-training network to fine-tune the entire pre-training network to complete the classification task. Given a set of input data, the probability that the input belongs to a certain class i is equal to

In the formula, R is the input of the logistic regression layer, W, b are the weight and bias of the logistic regression layer, and the sum of all outputs is 1.

The logistic regression layer and the stacked autoencoder form a deep classifier. The process of training the deep classifier is as follows:

In the present invention, two different pattern classifiers are built according to actual needs, which are recorded as SAE_NB and SAE_LM, and the above-mentioned training rules are used respectively to train two deep pattern classifiers in the selected representative training set. Use SAE_NB to divide the pixels into boundary pixels and non-boundary pixels, and then use SAE_LM to divide the boundary pixels into LII and MAI pixels.

3 boundary extraction

Due to the low resolution and signal-to-noise ratio of ultrasound images and the relatively high similarity of patterns, misclassification occurs. False positive errors, that is, wrongly classify non-boundary points as boundary points, have a great impact on the final result. Therefore, it is necessary to screen the classification results. The present invention uses the area information and position of the target classification area to select the classification area. Use the row index value of the farthest end of the carotid artery and the barycenter row value and area information of the area to be determined to remove misclassified areas that are close to the target area but have a large area, and use the center of gravity between the largest area of the target and the remaining smaller areas to be determined The row values are compared, combined with the area information of the smaller area to remove the area that is far away from the target area and the area is small, and finally indexed by column to remove the area with incomplete intima-media structure.

Due to the poor quality of the ultrasound images, the derived borders are not a single pixel wide, which is the main reason for interobserver variability in manual measurements. In order to eliminate this difference, according to the principle of the minimum sum of squared deviations, polynomial curve fitting is carried out on the classification boundary, and the process is as follows:

Let the fitting polynomial be:

f＝a ₀ +a ₁ e+...+a _k e ^k (21)

Among them, e and f are the column value and row value of the boundary point pixel respectively, ak and are the polynomial coefficients, and k is the degree of the polynomial.

(2) The sum of the distances from each point to this curve, that is, the square of the deviation is as follows:

Among them, n is the size of the data set, (ei, fi) is the column and row values of the i-th pixel, and R represents the deviation.

(3) Calculate the partial derivative of ai on the right side of the equation to get

(4) Simplify the left formula to get the following formula

(5) Express the equation in matrix form

(6) Simplify the Vandermonde matrix obtained in (5) to get

(7) (6) formula is E*A=F, then

A＝(E*E) ^-1 *E*F (27)

When the coefficient matrix is obtained, the fitting curve is obtained at the same time.

Beneficial effect

The invention can effectively solve the problem of generally poor overall performance in current IMT algorithms, and can realize fully automatic, fast, accurate and robust measurement of IMT.

In the ROI selection stage, the CNN is used to identify the distal end of the ultrasonic carotid artery. CNN has a certain tolerance to the input data with large deformation and has certain robustness to noise; Different images can accurately locate the distal end of the carotid artery and correctly extract the ROI.

In the pixel classification stage, according to the different characteristics of the classification set, different stacked autoencoders are built; the stacked autoencoder extracts the features of the input data layer by layer, and finally obtains the deep features of the data and classifies them to ensure the efficiency of classification. sex and effectiveness.

In the boundary extraction stage, considering the incomplete intima-media structure of the ultrasound images used, it is inevitable that there will be fractures. When selecting reliable classification regions, not only the area of the region but also the location of the region should be considered.

Fig. 3 describes a flow chart of an IMT measurement algorithm based on deep learning proposed by the present invention. Firstly, the input ultrasonic carotid artery image is cropped, averaged, and sub-images are obtained, and the sub-images are classified and predicted, and the row index value that can effectively calibrate the distal end of the carotid artery is obtained, and the ROI is obtained. Use a specific window to process the ROI, first use the window-processed pixel pattern information as the input of the SAE_NB classifier, classify the pixels into non-boundary pixels and boundary pixels, and then use the SAE_LM classifier to classify the boundary pixels into LII and MAI Pixel, and finally use the area information and position information of the classification area to select the classification results, search for the classification area containing the complete intima-media structure, and fit the final boundary according to the principle of the least square sum of deviations to complete the IMT measurement.

Claims (5)

1. a method for measuring carotid intima-media thickness based on deep learning, is characterized in that, the steps are as follows: 1) ROI acquisition: the convolutional neural network CNN (Convolution Neural Network) is used to automatically identify the distal end of the ultrasonic carotid artery, and then extract the ROI; 2) Pixel classification: The pixel gray value information of a certain area centered on the pixel to be classified is used as the input of the stacked autoencoder, and its output is used as the input for higher-level feature extraction. After the feature extraction, a logistic regression classification layer is added to construct a pattern classifier , to complete the classification of image pixels; 3) Boundary extraction: use the area information and position of the target classification area to select the classification area, and use the row index value of the farthest end of the carotid artery and the barycenter row value and area information of the area to be determined to remove the area that is close to the target area but has a smaller area. For large misclassified areas, use the largest area of the target to compare the barycenter row values of the remaining smaller areas to be determined, combine the area information of the smaller areas to remove areas that are farther away from the target area and have a smaller area, and finally index by column, Regions with incomplete intima-media structure were removed; polynomial curve fitting was performed on the classification boundary according to the principle of minimum sum of squared deviations. 2. the ultrasonography carotid artery intima-media thickness measurement method based on deep learning as claimed in claim 1, is characterized in that, adopts convolutional neural network (CNN) to extract ROI concrete steps are as follows: (1) Crop the image, and cut out the information irrelevant to the image analysis in the image; (2) the cropped image is evenly divided into five equal columns, and image blocks of a certain size are sequentially taken out along the symmetrical axis of the equally divided sub-image; (3) The image block is used as the input of the trained CNN for predictive classification, the number of categories is 2, that is, the image containing the "dark-bright-dark-bright" structure and the image without this structure, and the same sub-image is selected The middle attribute contains the image block with the largest predicted value of the "dark-bright-dark-bright" structure class; (4) Sorting the row index values of the obtained image blocks, selecting the row index values that can effectively calibrate the distal end of the carotid artery, and extracting the ROI based on the index values, the selection rules are as follows. r1, r2 and r3 are the maximum value, the second maximum value and the middle value of the row index value respectively. m1, m2 are the mean values of r1, r2 and r3, r2 respectively, v is a set threshold, and r3 is preferred as the effective index value of the distal end of the carotid artery. If r3 is less than a certain threshold, it is considered that r3 cannot effectively define the distal end of the carotid artery. At this time, use m1 or m2. 3. the ultrasonography carotid artery intima-media thickness measurement method based on deep learning as claimed in claim 1, is characterized in that, the algorithm execution process of self-encoding decoder AE (Auto Encoder) comprises encoding process and decoding process, and its operation process As shown in formula (1) (2): y＝f(w _y *x+b _y ) (1) z＝f(w _z *y+b _z ) (2) In the formula, x represents the input of AE, and also represents the output of the hidden layer, z represents the output of AE, w _y , b _y are the weights and biases from the input layer to the hidden layer, w _z , b _z are the weights and biases from the hidden layer to the hidden layer, respectively. The weight and bias of the output layer, f(.) is the activation function; The training process of AE is the process of minimizing the error function, $\underset{{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}}{\mathrm{<span\; class="notranslate">\; arg</span>}\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; \&lsqb;</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>$ In order to reduce the number of training parameters, usually set W _y =W _z =W (3) Choose the following rules to update the weight term and bias term to minimize the error function: $\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&eta;</span>}\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; w</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$ ${\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&eta;</span>}\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$ ${\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&eta;</span>}\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span>}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$ The stacked autoencoder SAE (Stacked Auto Encoder) is composed of a multi-layer autoencoder, the output of the previous layer of autoencoder is used as the input of the next layer of autoencoder, and the activation value vector of the deepest hidden layer unit is the input higher-order representation of values; Select the sigmoid function as the activation function of the codec. In the pre-training process, the network mainly learns three parameters W, b _y , b _z . Before setting the parameter update rules, it is necessary to reasonably determine the error term, that is, the cost function. Select Cross entropy is used as a cost function, and at the same time the entire training data is divided into small pieces of data, so for the entire data set, the parameters are updated with small pieces of data: $\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; e</span>}}^{\mathrm{<span\; class="notranslate">\; x</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$ $\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; d</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; l</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$ In the formula, d is the dimension of the input vector, m is the size of the data block, x _ik and z _ik are the kth component of the i-th input data in the selected block respectively; To optimize formula (8) with stochastic gradient descent method, first express the reconstruction layer in the form of scalar: ${\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; p</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; d</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; q</span>}}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$ ${\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; p</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$ ${\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; d</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; q</span>}}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; q</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}\mathrm{<span\; class="notranslate">\; p</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$ Among them, net _ip ^y is the p-th dimension of the hidden layer unit, netikz i is the p-th dimension of the output unit, which represents the number of the block data, d represents the dimension of the input data, h represents the dimension of the hidden layer, z _ik is the dimension of the input data k-th dimension reconstruction; The first and second derivatives of (7) are: f'(x)=f(x)1-f(x) (12) f”(x)=f(x)1-f(x)1-2f(x) (13) According to (9)-(11), calculate the partial derivative of the reconstruction layer with respect to w, b _y , b _z ; $\frac{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span>\\ {\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; \&NotEqual;</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; )</span>\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$ $\frac{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; r</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; r</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; )</span>$ $\frac{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; )</span>$ In the formula, Wrs is the weight connecting the rth input and the sth hidden layer unit, b _yr is the bias of the rth unit in the hidden layer, and b _zr is the bias of the rth unit in the reconstruction layer; The partial derivatives of the cost function with respect to W, b _y , b _z are obtained from equations (9)-(16): $\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; c</span>}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; \{</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; d</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; \&lsqb;</span>\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \}</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 17</span>}<span\; class="notranslate">\; )</span>$ $\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; c</span>}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; the\; y</span>}\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; d</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; r</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; r</span>}}^{\mathrm{<span\; class="notranslate">\; the\; y</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 18</span>}<span\; class="notranslate">\; )</span>$ $\frac{<span\; class="notranslate">\; \&part;</span>\mathrm{<span\; class="notranslate">\; c</span>}}{<span\; class="notranslate">\; \&part;</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; z</span>}\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; d</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; z</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; f</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; net</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; z</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 19</span>}<span\; class="notranslate">\; )</span>$ After the network training is completed, the data reconstruction of the network is removed, the output of the hidden layer is the learned feature, and the output of the subsequent network layer is the input of the previous layer, and it is trained in the same way, that is, the network is greedily trained layer by layer, so as to avoid the loss of the initial weight of the network. The value is too small to fall into the local optimal solution; finally, combine the layers of the network together, use the features learned by the network to classify, and add a soft-max classification layer after the pre-training network to fine-tune the entire pre-training network to complete the classification task. Given a set of input data, the probability that the input belongs to a certain class i is equal to $\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; Y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; W</span>}\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}}{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{{\mathrm{<span\; class="notranslate">\; e</span>}}^{{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; b</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 20</span>}<span\; class="notranslate">\; )</span>$ In the formula, R is the input of the logistic regression layer, W, b are the weight and bias of the logistic regression layer, and the sum of all outputs is 1. 4. as claimed in claim 3, based on the deep learning method for measuring the thickness of the supersonic carotid artery intima-media, it is characterized in that the logistic regression layer and the stacked autoencoder constitute the depth classifier, and the process of training the depth classifier is as follows: (1). Start (2). Initialize the data block size b of the block data, the number of pre-training pt, the pre-training learning rate pl, the number of fine-tuning (fine-tuning) times ft and the learning rate fl, the number of hidden layers d, the neural network of each hidden layer The number of elements [d], the dimension of the input data D, the number of classification categories C; (3) for each layer L (1≤L≤d) (4) Construct the input neurons of d-vis and the hidden layer neurons of d-hid (5) if L is the first layer, L=1 (6) d-vis=D (7) d-hid=n[1] (8) At this time, the original data is the input of AE (9)else (10) d-vis=n[L-1] (11)d-hid=n[L] (12) At this time, the input of AE is the output of the previous AE (13) end (14) Initialize the weight matrix W of AE with random variables, by, bz are initialized to zero (15) for each pre-training (16) for each data block (17) Calculate the reconstruction layer: z=f(W·f(W·x+b _y )+b _z ) (18) Calculate the cost function: $\mathrm{<span\; class="notranslate">\; c</span>}<span\; class="notranslate">\; =</span><span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; b</span>}}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; x</span>}\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; log</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; z</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>$ (19) Utilize formula (17)-(19) to update the weight, and the learning rate is pl (20) end (21) end (22) Remove the data reconstruction layer (23) end (24) Initialize the logistic regression layer, the number of input neurons is n[d], and the number of output neurons is C (25) for every fine adjustment (26) for each data block (27) Calculate the possibility of each category by formula (20) (28) Use the classic backpropagation method to update the weight and bias items of the network, and the learning rate is fl (29) end (30) end (31) end Build two different pattern classifiers according to actual needs, denoted as SAE_NB and SAE_LM, respectively use the above training rules to train two deep pattern classifiers in the selected representative training set, and use SAE_NB to divide pixels into boundaries pixels and non-boundary pixels, and then use SAE_LM to divide the boundary pixels into LII and MAI pixels. According to the principle of the minimum sum of squared deviations, polynomial curve fitting is performed on the classification boundary, and the process is as follows: Let the fitting polynomial be: f＝a ₀ +a ₁ e+...+a _k e ^k (21) Among them, e and f are the column value and row value of the boundary point pixel respectively, a _k is the polynomial coefficient, and k is the polynomial degree. (2) The sum of the distances from each point to this curve, that is, the square of the deviation is as follows: ${\mathrm{<span\; class="notranslate">\; R</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; twenty\; two</span>}<span\; class="notranslate">\; )</span>$ Among them, n is the size of the data set, (e _i , f _i ) is the column and row values of the i-th pixel, and R represents the deviation. (3) Calculate the partial derivative of a _i on the right side of the equation to get $<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}$ $<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}$ ... $<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&rsqb;</span><span\; class="notranslate">\; \&Center\; Dot;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; twenty\; three</span>}<span\; class="notranslate">\; )</span>$ (4) Simplify the left formula to get the following formula ${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}$ ${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}$ ... ${\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; twenty\; four</span>}<span\; class="notranslate">\; )</span>$ (5) Express the equation in matrix form $\left[\begin{array}{cccc}\mathrm{<span\; class="notranslate">\; no</span>}& {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}& {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\\ <span\; class="notranslate">\; ...</span>& <span\; class="notranslate">\; ...</span>& <span\; class="notranslate">\; ...</span>& <span\; class="notranslate">\; ...</span>\\ {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}}& {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}& <span\; class="notranslate">\; ...</span>& {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right]\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\\ {\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ <span\; class="notranslate">\; ...</span>\\ {\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\\ {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\\ <span\; class="notranslate">\; ...</span>\\ {\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 25</span>}<span\; class="notranslate">\; )</span>$ (6) Simplify the Vandermonde matrix obtained in (5) to get $\left[\begin{array}{cccc}\mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}& <span\; class="notranslate">\; ...</span>& {{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}\\ \mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}& <span\; class="notranslate">\; ...</span>& {{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}\\ <span\; class="notranslate">\; ...</span>& <span\; class="notranslate">\; ...</span>& <span\; class="notranslate">\; ...</span>& <span\; class="notranslate">\; ...</span>\\ \mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}& <span\; class="notranslate">\; ...</span>& {{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}}^{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right]\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\\ {\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ <span\; class="notranslate">\; ...</span>\\ {\mathrm{<span\; class="notranslate">\; a</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\end{array}\right]<span\; class="notranslate">\; =</span>\left[\begin{array}{c}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}\\ {\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ <span\; class="notranslate">\; ...</span>\\ {\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}}\end{array}\right]<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 26</span>}<span\; class="notranslate">\; )</span>$ (7) (6) formula is E*A=F, then A＝(E*E) ^-1 *E*F (27) When the coefficient matrix is obtained, the fitting curve is obtained at the same time. 5. An ultrasonic carotid artery intima-media thickness measuring device based on deep learning is characterized in that it comprises an ultrasonograph and a computer, and the image signal generated by the ultrasonograph is processed by a computer, and the computer is provided with the following modules to process ultrasonic images : 1) ROI acquisition module: the convolutional neural network CNN (Convolution Neural Network) is used to automatically identify the distal end of the ultrasonic carotid artery, and then extract the ROI; 2) Pixel classification module: The pixel gray value information of a certain area centered on the pixel to be classified is used as the input of the stacked autoencoder, and its output is used as the input for higher-level feature extraction. After the feature extraction, a logistic regression classification layer is added to construct the pattern classification device to complete the classification of image pixels; 3) Boundary extraction module: use the area information and position of the target classification area to select the classification area. Use the row index value of the farthest end of the carotid artery and the barycenter row value and area information of the area to be determined to remove misclassified areas that are close to the target area but have a large area, and use the barycenter of the largest area of the target and the remaining smaller areas to be determined Compare the row values, combine the area information of the smaller area to remove the area that is farther away from the target area and have a smaller area, and finally index by column to remove the area with incomplete intima-media structure; Perform a polynomial curve fit.