CN104574408A - Industry transparent film package detecting method and device based on shape feature extraction - Google Patents

Abstract

The invention relates to an industrial transparent film packaging detection method and device based on shape feature extraction. The method includes: preprocessing the obtained packaging product image; performing superpixel segmentation on the preprocessed image; using seven Hu invariant moments to define shape features to realize shape feature extraction, and introducing new parameters; The matrix is processed to obtain the principal components; the least squares support vector machine is used to train according to the shape features, and then the superpixel blocks are classified. The device includes a conveyor belt module, a picture shooting module, an algorithm module and a result analysis module. The invention can reduce detection time while maintaining high classification performance.

Description

Detection method and device for industrial transparent film packaging based on shape feature extraction

technical field

The invention relates to the technical field of transparent film packaging detection, in particular to an industrial transparent film packaging detection method and device based on shape feature extraction.

Background technique

The outer packaging film of the product often has defects such as packaging bubbles, wrinkles, and slack, which seriously affect the image of the product. Due to the high light transmittance of the packaging film, manual visual inspection is mostly used at present, and the detection effect is not ideal. The experience and skills of the operator are greatly affected. It is impossible to automatically detect the heat-sealing defects of the outer packaging film of the product on the automatic production line.

Contents of the invention

The technical problem to be solved by the present invention is to provide an industrial transparent film packaging detection method and device based on shape feature extraction, which can reduce the detection time while maintaining high classification performance.

The technical solution adopted by the present invention to solve the technical problem is to provide a detection method for industrial transparent film packaging based on shape feature extraction, comprising the following steps:

(1) Preprocessing the obtained packaging product image;

(2) Carry out superpixel segmentation to the preprocessed image;

(3) Use seven Hu invariant moments to define shape features to realize shape feature extraction, and introduce new parameters;

(4) Process the multivariate parameter matrix to obtain the principal components;

(5) Use the least squares support vector machine to train according to the shape features, and then classify the superpixel blocks.

The image of the packaged product is captured using a bowl-shaped light source and a wide-angle lens.

In the step (1), the Canny edge detection is used first, and then the expansion operation is used to preprocess the obtained packaging product image.

The step (2) adopts a simple linear iterative clustering method to perform superpixel segmentation.

In the described step (3), seven Hu invariant moments are respectively:

φ ₁ =η ₂₀ +η ₀₂ ;

${\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 20</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 02</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 4</span>}{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ;</span>$

φ ₃ =(η ₃₀ -3η ₁₂ ) ² +(3η ₂₁ -η ₀₃ ) ² ;

φ ₄ =(η ₃₀ +η ₁₂ ) ² +(η ₂₁ +η ₀₃ ) ² ;

φ ₅ =(η ₃₀ -3η ₁₂ )(η ₃₀ +η ₁₂ )[(η ₃₀ +η ₁₂ ) ² -3(η ₂₁ +η ₀₃ ) ² ]

+(3η ₂₁ -η ₀₃ )(η ₂₁ +η ₀₃ )[3(η ₃₀ +η ₁₂ ) ² -(η ₂₁ +η ₀₃ ) ² ];

φ ₆ =(η ₂₀ -η ₀₂ )[(η ₃₀ +η ₁₂ ) ² -(η ₂₁ +η ₀₃ ) ² ]+4η ₁₁ (η ₃₀ +η ₁₂ )(η ₂₁ +η ₀₃ );

φ ₇ ＝(3η ₂₁ -η ₀₃ )(η ₃₀ +η ₁₂ )[(η ₃₀ +η ₁₂ ) ² -3(η ₂₁ +η ₀₃ ) ² ]+

(η ₀₃ -η ₁₂ )(η ₂₁ +η ₀₃ )[3(η ₃₀ +η ₁₂ ) ² -(η ₂₁ +η ₀₃ ) ² ];

η _ij represents the (i+j) order normalized central moment of the image. The central moment of (i+j) of f(x+y) of the image function is defined as Where Ω is the value interval of x and y. For N*M digital images, using summation instead of integral, the (i+j) order center can be expressed as Then the (i+j) order formatted central moment can be expressed as ${\mathrm{\&eta;}}_{\mathrm{ij}}=\frac{{\mathrm{\&mu;}}_{\mathrm{ij}}}{{{\mathrm{\&mu;}}^{\mathrm{\&gamma;}}}_{00}},$ in $\mathrm{\&gamma;}=\frac{i+j}{2}+1(i+j=\mathrm{2,3},...);$

Among them, the normalized moments of the seven Hu invariant moments are invariant to translation, scaling, stretching and extrusion; the normalized central moments of the first six Hu invariant moments are invariant to rotation; the seventh Hu invariant The normalized central moments of moments are invariant to rotation and invariant to twist.

The new parameters introduced in the step (3) include: area, perimeter, compactness, number of holes, ratio of number of holes and area; wherein, area: used to calculate the number of pixels contained in holes; perimeter: the number of holes Measured by the sum of the distances between pixels on the contour line; compactness: Among them, S is the area, L is the perimeter; the number of holes: the number of holes on a package; the ratio of the number of holes to the area: used to distinguish large holes from small holes.

In the step (4), use the least squares support vector machine to select 1000 holes and 2000 non-holes for training, and use the trained classifier to classify the segmented superpixels.

The technical solution adopted by the present invention to solve its technical problems is: it also provides an industrial transparent film packaging detection device based on shape feature extraction, including: a conveyor belt module, used to transport packaged products; a picture shooting module, located directly above the conveyor belt module , used to acquire the image of the conveying packaged product; the algorithm module is connected with the picture shooting module, and is used to process the picture according to the above-mentioned detection method; the result analysis module is used to analyze the result of the picture processing.

Due to the adoption of the above technical solution, the present invention has the following advantages and positive effects compared with the prior art: the shape feature extraction adopted by the present invention not only utilizes seven Hu invariant moments to define shape features, but also introduces new Parameters, dimensionality reduction is also performed by principal component analysis, and LSSVM is used for training, and then the trained classifier is used to classify superpixel blocks. The method of the invention not only reduces the time, but also has better effect and is industrially feasible.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Fig. 2 is a schematic diagram of the structure of the device using the present invention.

Detailed ways

Below in conjunction with specific embodiment, further illustrate the present invention. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the teachings of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

Embodiments of the present invention relate to an industrial transparent film packaging detection method based on shape feature extraction, as shown in Figure 1, comprising the following steps:

(1) Preprocessing the obtained packaging product image;

(2) Carry out superpixel segmentation to the preprocessed image;

(3) Use seven Hu invariant moments to define shape features to realize shape feature extraction, and introduce new parameters;

(4) Process the multivariate parameter matrix to obtain the principal components;

(5) Use the least squares support vector machine to train according to the shape features, and then classify the superpixel blocks.

Each step is described in detail below.

Preprocessing:

First use Canny edge detection, and then use expansion operation.

Superpixel segmentation:

In the past, the understanding of images is a two-dimensional matrix composed of pixels, so the segmentation is also based on pixels, but the pixel-based segmentation will lead to low processing efficiency, so the present invention uses superpixel segmentation. A superpixel is a collection of connected, similar in brightness or color pixels in a local area of an image. What the present invention adopts is SLIC (Simple Linear Iterative Clustering) superpixel segmentation.

feature extraction

The present invention uses seven Hu invariant moments to define shape features, and also introduces new parameters. The expressions of the seven Hu invariant moments are as follows:

φ ₁ =η ₂₀ +η ₀₂

${\mathrm{<span\; class="notranslate">\; \&phi;</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 20</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 02</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 4</span>}{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}_{\mathrm{<span\; class="notranslate">\; 11</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$

φ ₃ ＝(η ₃₀ -3η ₁₂ ) ² +(3η ₂₁ -η ₀₃ ) ²

φ ₄ ＝(η ₃₀ +η ₁₂ ) ² +(η ₂₁ +η ₀₃ ) ²

φ ₅ ＝(η ₃₀ -3η ₁₂ )(η ₃₀ +η ₁₂ )[(η ₃₀ +η ₁₂ ) ² -3(η ₂₁ +η ₀₃ ) ² ]+

(3η ₂₁ -η ₀₃ )(η ₂₁ +η ₀₃ )[3(η ₃₀ +η ₁₂ ) ² -(η ₂₁ +η ₀₃ ) ² ]

φ ₆ =(η ₂₀ -η ₀₂ )[(η ₃₀ +η ₁₂ ) ² -(η ₂₁ +η ₀₃ ) ² ]+4η ₁₁ (η ₃₀ +η ₁₂ )(η ₂₁ +η ₀₃ )

φ ₇ ＝(3η ₂₁ -η ₀₃ )(η ₃₀ +η ₁₂ )[(η ₃₀ +η ₁₂ ) ² -3(η ₂₁ +η ₀₃ ) ² ]+

(η ₀₃ -η ₁₂ )(η ₂₁ +η ₀₃ )[3(η ₃₀ +η ₁₂ ) ² -(η ₂₁ +η ₀₃ ) ² ]

Among them, η _ij represents the (i+j) order normalized central moment of the image. The central moment of (i+j) of f(x+y) of the image function is defined as Where Ω is the value interval of x and y. For N*M digital images, using summation instead of integral, the (i+j) order center can be expressed as Then the (i+j) order formatted central moment can be expressed as ${\mathrm{\&eta;}}_{\mathrm{ij}}=\frac{{\mathrm{\&mu;}}_{\mathrm{ij}}}{{{\mathrm{\&mu;}}^{\mathrm{\&gamma;}}}_{00}},$ in $\mathrm{\&gamma;}=\frac{i+j}{2}+1(i+j=\mathrm{2,3},...).$

The calculated invariant moment features are F _m = φ ₁ , φ ₂ ... φ ₇ , where the value of the high-order moments is very small, so it needs to be standardized when matching, and the normalized moments Invariant to translation, scaling, stretching and pinching changes. In addition, the first 6 normalized central moments are invariant to rotation, while the seventh is also invariant to twist.

Although these seven moments can describe the shape features well, when the image database is large, only these seven scalars are not enough. The present invention introduces new parameters: area, perimeter, density, number of holes, The ratio of the number of holes to the area. The expressions of area, perimeter, density, number of holes, and ratio of number of holes to area are as follows:

1) Area: used to calculate the number of pixels contained in the hole;

2) Circumference: Measured by the sum of the distances between pixels on the contour line of the hole, the distance between parallel pixels is 1 pixel, and the distance between pixels in the oblique direction needs to be measured according to the connection between pixels. method to calculate the distance separately;

3) Density: Where S is the area and L is the perimeter;

4) Number of holes: calculate the number of holes on a package;

5) The ratio of the number of holes to the area: it is mainly used to distinguish large holes from small holes.

principal component analysis

The main idea of principal component analysis is to perform matrix processing on the multivariate parameter matrix, and obtain a linear combination of the original variables, which are not correlated with each other, and can reflect the information contained in the original variables to the maximum extent.

LSSVM classification

The segmented superpixels are classified into holes and non-holes using Least Squares Support Vector Machine (LSSVM). For the least squares support vector machine, the optimization problem can be expressed as:

$\mathrm{<span\; class="notranslate">\; min</span>}\mathrm{<span\; class="notranslate">\; J</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; \&xi;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&gamma;</span>}\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; t</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&xi;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4.9</span>}<span\; class="notranslate">\; )</span>$

Using Lagrangian to solve the above optimization problem is transformed into solving a linear equation problem.

In the training of LSSVM, 1000 holes and 2000 non-holes are selected for training, that is, 3000 sets of data are trained. The segmented superpixels are classified using the trained classifier, and each superpixel is divided into holes and non-holes.

The second embodiment of the present invention relates to an industrial transparent film packaging detection device based on shape feature extraction, as shown in FIG. The conveyor belt module 1 is composed of a conveyor belt, and its function is to convey industrial packaging products at a constant speed. The picture shooting module 2 is located directly above the conveyor belt module, and its function is to use a bowl-shaped light source and a wide-angle lens to capture images of the packaging products. The function of the algorithm module 3 is based on The above detection method performs image processing, and the function of the result analysis module 4 is to analyze the image processing result.

It is not difficult to find that the shape feature extraction adopted in the present invention not only uses seven Hu invariant moments to define shape features, but also introduces new parameters, and also uses principal component analysis to perform dimension reduction processing, and uses LSSVM for training, The trained classifier is then used to classify superpixel patches. The method of the invention not only reduces the time, but also has better effect and is industrially feasible.

Claims (8)

1. an industrial transparent film packaging detection method for Shape-based interpolation feature extraction, is characterized in that, comprise the following steps:

(1) pre-service is carried out to the packaging product image obtained;

(2) pretreated image is carried out to the segmentation of super-pixel;

(3) utilize seven Hu not bending moment definition shape facility realize Shape Feature Extraction, and introduce new parameter;

(4) multivariable parameter matrix is processed, obtain major component;

(5) use least square method supporting vector machine to train according to shape facility, then super-pixel block is classified.

2. the industrial transparent film packaging detection method of Shape-based interpolation feature extraction according to claim 1, is characterized in that, described packaging product image uses bowl-shape light source and wide-angle lens to obtain.

3. the industrial transparent film packaging detection method of Shape-based interpolation feature extraction according to claim 1, it is characterized in that, in described step (1), employing first utilizes Canny rim detection, and recycling expansive working carries out pre-service to the packaging product image obtained.

4. the industrial transparent film packaging detection method of Shape-based interpolation feature extraction according to claim 1, is characterized in that, described step (2) adopts the mode of simple linear iteration cluster to carry out super-pixel segmentation.

5. the industrial transparent film packaging detection method of Shape-based interpolation feature extraction according to claim 1, is characterized in that, in described step (3) seven Hu not bending moment be respectively:

φ ₁＝η ₂₀+η ₀₂；

φ ₂＝(η ₂₀-η ₀₂) ²+4η ₁ ² ₁；

φ ₃＝(η ₃₀-3η ₁₂) ²+(3η ₂₁-η ₀₃) ²；

φ ₄＝(η ₃₀+η ₁₂) ²+(η ₂₁+η ₀₃) ²；

φ ₅＝(η ₃₀-3η ₁₂)(η ₃₀+η ₁₂)[(η ₃₀+η ₁₂) ²-3(η ₂₁+η ₀₃) ²]

+(3η ₂₁-η ₀₃)(η ₂₁+η ₀₃)[3(η ₃₀+η ₁₂) ²-(η ₂₁+η ₀₃) ²]；

φ ₆＝(η ₂₀-η ₀₂)[(η ₃₀+η ₁₂) ²-(η ₂₁+η ₀₃) ²]+4η ₁₁(η ₃₀+η ₁₂)(η ₂₁+η ₀₃)；

φ ₇＝(3η ₂₁-η ₀₃)(η ₃₀+η ₁₂)[(η ₃₀+η ₁₂) ²-3(η ₂₁+η ₀₃) ²]+

(η ₀₃-η ₁₂)(η ₂₁+η ₀₃)[3(η ₃₀+η ₁₂) ²-(η ₂₁+η ₀₃) ²]；

η _ijrepresent normalization center, (i+j) rank square of image;

Wherein, the normalized moments of seven Hu not bending moment is constant to translation, convergent-divergent, stretching, extension and extruding change; The normalization center square of the first six Hu not bending moment is to invariable rotary; The normalization center square of the 7th Hu not bending moment is also constant to distortion to invariable rotary.

6. the industrial transparent film packaging detection method of Shape-based interpolation feature extraction according to claim 1, it is characterized in that, introduce new parameter in described step (3) and comprise: area, girth, density, hole number, hole number and area ratio; Wherein, area: be used for calculating the pixel count that comprises of hole; Girth: on the outline line of hole, pel spacing is measured from sum; Density: wherein S is area, and L is girth; Hole number: the hole number in a packaging; Hole number and area ratio: be used for distinguishing macroscopic void and small holes.

7. the industrial transparent film packaging detection method of Shape-based interpolation feature extraction according to claim 1, it is characterized in that, described step (4) middle use least square method supporting vector machine chooses 1000 holes and 2000 non-holes are trained, and uses the sorter of training out to classify to the super-pixel split.

8. an industrial transparent film packaging pick-up unit for Shape-based interpolation feature extraction, is characterized in that, comprising: conveyor belt module, for transmitting packaging product; Picture shooting module, is positioned at directly over conveyor belt module, for obtaining the image transmitting packaging product; Algoritic module, is connected with described picture shooting module, for carrying out picture processing according to the detection method as described in claim arbitrary in claim 1-7; Results analyses module, for analyzing the result of picture processing.