CN118429817A - Intelligent analysis method of substation site map change area based on image comparison - Google Patents

Abstract

The invention relates to the technical field of intelligent patrol of transformer substations, in particular to an intelligent analysis method for a change area of a bitmap of a transformer substation based on image comparison, which is used for comparing a reference map with a bitmap of a corresponding point and comprises the following steps: image preprocessing, screening according to color characteristics, analyzing an image change region, and finally framing a change region on a changed point bitmap to give an alarm. The invention does not need to train a deployment model for a long time based on a deep learning frame, can be directly deployed and used, occupies less hardware resources such as video memory, memory and the like, and saves cost; the change areas are screened through two color channel chart algorithms, and meanwhile, a complex change analysis algorithm aiming at different conditions is added, so that the screening accuracy is improved.

Description

Intelligent analysis method of substation site map change area based on image comparison

Technical Field

The present invention relates to the technical field of intelligent inspection of substations, and in particular to an intelligent analysis method of substation site bitmap change areas based on image comparison.

Background technique

Due to the long operation and maintenance time and power outages, when executing the inspection task, the equipment in the same point image of the current task may change compared with the reference map of the point in the reference point library due to changes in equipment position, foreign body intrusion, etc. These changes do not belong to typical defects of substations. The existing target detection and recognition algorithm cannot analyze the changes in the point map, and failure to discover them in time will affect the actual operation and maintenance effect. The invention patent with the authorization announcement number CN113724247B discloses a substation intelligent inspection method based on image discrimination technology, which combines the heat map generation in the visual field with the semantic segmentation in the deep learning field to perform intelligent detection and analysis on the real-time picture to determine whether there is an abnormality in the picture. However, this technical solution requires the preparation of training data in advance, and can only be deployed and used after long-term training under the deep learning framework, which occupies a large amount of video memory and memory resources; in addition, the invention only uses a single image for direct comparison and difference, plus simple binarization and noise reduction processing, which may result in a relatively low discrimination accuracy.

Summary of the invention

In order to overcome the shortcomings of the prior art, the present invention provides an intelligent analysis method for substation point map change areas based on image comparison. Through a series of change area analysis algorithms, the interference of background factors outside the equipment is eliminated, and the area where the point map has changed compared with the reference map is accurately identified.

To achieve the above-mentioned purpose, the present invention discloses an intelligent analysis method for substation site bitmap change area based on image comparison, which includes determining a number of patrol points in a patrol area, encoding each point, and setting a shooting device at the patrol point; establishing a reference library of each point in the patrol area; when performing a patrol task, using a shooting device to capture a point map of the patrol point; extracting a reference map in the reference library according to the point code and performing image comparison with the corresponding point map; specifically including the following steps:

S1: image preprocessing, which includes image downsampling, key point screening, and point map offset correction;

S2: screening according to color features, wherein the screening according to color features includes subtracting RGB color channels, subtracting the maximum value and average value of RGB color channels, and a difference map selection algorithm;

S3: image change region analysis, including corrosion, seasonal background change elimination, expansion light shadow elimination, connected domain filtering, and change frame merging and restoration;

S4: Frame the changed area on the point map where the change occurs and issue an alarm.

Furthermore, the image downsampling includes compressing the aspect ratio of the reference image to 1/4 of the original aspect ratio to obtain the reference image img1; the image downsampling also includes compressing the aspect ratio of the point map to 1/4 of the original aspect ratio to obtain the point map img2.

Furthermore, the key point screening steps are as follows:

S11: Using the SIFT-based feature point matching algorithm, feature points of the reference image img1 and the point image img2 are detected, and feature descriptors are calculated for each feature point;

S12: The nearest neighbor relationship between the feature descriptors of the reference image img1 and the point image img2 is identified through the brute force matching method. In order to ensure the accuracy of the matching, the K nearest neighbor matching strategy is introduced to determine the best matching feature points and the second best matching feature points;

S13: Setting a distance ratio filtering threshold, and using the distance ratios between the best matching feature point, the second best matching feature point and the corresponding feature point, to complete the key point pair screening.

Furthermore, the point map offset correction includes: if the number of screened key point pairs exceeds 4, the optimal single mapping transformation matrix between these key point pairs is calculated, and the point map is restored and aligned in the X and Y directions based on this, to generate a registration map img_s that is spatially aligned with the reference map img1; if the number of retained feature point pairs is less than 4, the point map img2 is directly used as the registration map img_s.

Furthermore, the screening according to the color characteristics comprises the following steps:

S21: Directly subtract the reference image img1 from the registration image img_s to obtain the difference image img_diff;

S22: The RGB color channel difference includes separating the R, G, and B channels of the difference image img_diff, and then calculating the absolute differences diff_R_G, diff_G_B, and diff_B_R between the red and green, green and blue, and blue and red channels, and creating a mask for each difference; setting the threshold range 1 to 35 to 255, and the pixel values within the threshold range 1 are retained in the mask, and then merging the three masks to obtain: a channel difference binary image diff_RGB;

S23: Subtracting the maximum and average values of the RGB color channels includes: separating the R, G, and B channels of the difference image img_diff, first taking the average values R_mean, G_mean, and B_mean of the three channels respectively, and then calculating the absolute differences R-R_mean, G-G_mean, and B-B_mean between R, G, and B and the average values of the respective channels; and then taking the maximum values R_max, G_max, and B_max of the R, G, and B channels respectively, according to the following formula: , get the minimum threshold rgb_scale for pixel retention; finally, create masks for R-R_mean, G-G_mean, and B-B_mean of each channel, set threshold range 2 to rgb_scale to 255, and the pixels within the threshold range 2 are retained in the mask. Then merge the three masks to get: channel maximum value average difference binary map thresh_RGB.

Furthermore, a difference map selection algorithm is designed according to the color characteristics of the device in the scene image to select the difference binary map img_new. The difference map selection algorithm includes calculating the number of white pixels in the channel difference binary map diff_RGB and the channel maximum value average value difference binary map thresh_RGB And the proportion of white pixels to the entire image size , the specific algorithm selection logic is as follows:

If a1=0 and b2>0.5, select diff_RGB as the difference binary image img_new;

If a2=0 and b1>0.5, select thresh_RGB as the difference binary image img_new;

If a1=0 and a2>0, select thresh_RGB as the difference binary image img_new;

If a2=0 and a1>0, select diff_RGB as the difference binary image img_new.

Furthermore, the seasonal background change elimination includes: finding all the point-like connected domains in the difference binary image img_new, counting the number of the point-like connected domains, if the number of the point-like connected domains is greater than 30, and the area of each connected domain is less than 200 pixels, then filling these point-like connected domains as the background to obtain the difference binary image img_new1; performing image erosion processing on the difference binary image img_new1 to obtain the difference binary image img_erode.

Furthermore, the light shadow removal includes counting the number of connected domains in the difference binary image img_erode and the difference binary image img_new1, and calculating the ratio of the total number of pixels of all connected domains in the two images to the total circumscribed rectangular frame area of all connected domains. If the ratio is less than 0.4, and the number of connected domains in the difference binary image img_new1 is greater than 20 and the number of connected domains in the difference binary image img_erode is less than 3, then the connected domains in the img_erode image are filled as the background to obtain the difference binary image img_new2; and image dilation processing is performed on the difference binary image img_new2 to obtain the difference binary image img_dilate.

Furthermore, the connected domain filtering includes filling the connected domains with an area less than 200 pixels and a duty cycle less than 0.2 and the line segment-like connected domains with an aspect ratio greater than 5 as the background to obtain a difference binary image img_new3.

Furthermore, the change frame merging and restoring includes merging the external rectangular frames corresponding to all the remaining connected domains of the difference binary image img_new3 into one change frame, taking the four maximum values of the coordinates of all the external rectangular frames: x _min , x _max , y _min , y _max , and obtaining the output change frame coordinates ((x _min , y _min ), (x _max , y _max )); expanding the output change frame by 4 times to 1920 The changing box coordinates in 1080 pixels.

The present invention has at least the following beneficial effects:

The present invention uses a series of algorithms for screening and image change area analysis based on color features. It does not need to be based on a deep learning framework and does not require long-term training and deployment of models. It can be directly deployed and used to achieve accurate analysis of substation location image change areas. The change area intelligent recognition algorithm of the present invention has a fast analysis speed when actually deployed and applied, occupies less hardware resources such as video memory and internal memory, and saves costs. The present invention uses two color channel map algorithms to screen change areas based on the image features of equipment in the substation field, and at the same time adds complex change analysis algorithms for different situations, thereby improving the accuracy of change area recognition.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of the technical solution of the present invention.

Detailed ways

The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

The method for intelligent analysis of substation site bitmap change areas based on image comparison includes determining several patrol points in the patrol area, encoding each point, and setting up a shooting device at the patrol point; shooting a reference map of each point under normal conditions, and establishing a reference map library of each point in the patrol area; when performing patrol tasks, using a shooting device to capture the point map of the patrol point; extracting the reference map in the reference map library according to the point code and performing image comparison with the corresponding point map. Specifically, the following steps are included:

S1: Image preprocessing. The image preprocessing steps include image downsampling, key point screening, and point map offset correction.

The image downsampling specifically includes the following steps:

S10: Image downsampling algorithm is used to convert images to 1920 The 1080-resolution image is downsampled 4 times, that is, the width-to-height ratio of the reference image is compressed to 1/4 of the original, to obtain the reference image img1, and the width-to-height ratio of the point map is compressed to 1/4 of the original, to obtain the point map img2, so as to improve the image analysis speed and avoid the accumulation of point tasks caused by the long image analysis time.

Key point screening is based on feature point matching, which includes the following steps:

S11: Using the SIFT-based feature point matching algorithm, feature point detection is performed on the reference image img1 and the point image img2, and a feature descriptor is calculated for each feature point.

Feature points are salient points in an image. They remain stable under certain image transformations (such as rotation, scaling, and partial occlusion), so they can be used to match objects or scenes in different images. Feature descriptors are quantitative descriptions of the local area around a feature point, which captures specific properties of the area and is used to uniquely identify the feature point. By combining feature points and descriptors, feature point pairs that match across images can be found.

S12: The nearest neighbor relationship between the feature descriptors of the reference image img1 and the point image img2 is identified through the brute force matching method. In order to ensure the accuracy of the matching, the K nearest neighbor matching strategy is introduced to determine the best matching feature points and the second best matching feature points.

The best matching feature point is determined by using the nearest neighbor search. For each feature point in the reference image img1, the feature point in the point map img2 that is closest to the descriptor of the reference image img1 is found as the best matching feature point. Similarly, the second-best matching feature point is determined by using the next-best neighbor search. For each feature point in the reference image img1, the feature point in the point map img2 that is second closest to the descriptor of the reference image img1 is found as the second-best matching feature point.

S13: Setting a distance ratio filtering threshold, and using the distance ratios between the best matching feature point, the second best matching feature point and the corresponding feature point, to complete the screening of key point pairs in the point map img2 and the reference map img1.

The method is that when the distance ratio between the best matching feature point and the second best matching feature point and the corresponding feature point is less than 0.65, it means that these feature points have high similarity and correlation, so these feature points are retained as key points to complete the screening of key point pairs.

The point map offset correction specifically includes the following steps:

S14: If the number of the selected key point pairs exceeds 4, the optimal single mapping transformation matrix between these key point pairs is calculated, and the point map is restored and aligned in the X and Y directions based on it to generate a registration map img_s that is spatially aligned with the reference map img1; if the number of retained feature point pairs is less than 4, the point map img2 is directly used as the registration map img_s.

S2: Screening based on color features, the screening based on color features step includes the steps of subtracting the RGB color channels and subtracting the maximum and average values of the RGB color channels. In this step,

S21: Directly subtract the reference image img1 from the registration image img_s to obtain a difference image img_diff.

The subtraction of RGB color channels specifically includes the following steps:

S22: Separate the R, G, and B channels of the difference image img_diff, and then calculate the absolute differences diff_R_G, diff_G_B, and diff_B_R between the red and green, green and blue, and blue and red channels, and create a mask for each difference; set the threshold range one to 35 to 255, and the pixel values within the threshold range one are retained in the mask, and then merge the three masks to obtain: channel difference binary image diff_RGB.

The subtraction of the maximum value and the average value of the RGB color channel specifically includes the following steps:

S23: Separate the R, G, and B channels of the difference image img_diff, take the average values R_mean, G_mean, and B_mean of the three channels respectively, and then calculate the absolute differences R-R_mean, G-G_mean, and B-B_mean between R, G, and B and the average values of their respective channels; then take the maximum values R_max, G_max, and B_max of the R, G, and B channels respectively, according to the following formula: , respectively make the difference between the maximum value and the average value, and according to the scale factor of 0.4, get the minimum threshold rgb_scale for pixel retention; finally, create a mask for each channel's R-R_mean, G-G_mean, and B-B_mean, set the threshold range 2 to rgb_scale to 255, and the pixels within the threshold range 2 are retained in the mask. Then merge the three masks to get: channel maximum value average difference binary map thresh_RGB.

Finally, a difference image selection algorithm is designed according to the color features of the device in the scene image to select the difference binary image img_new, which specifically includes the following steps:

S24: Calculate the number of white pixels in the channel difference binary image diff_RGB and the channel maximum value average difference binary image thresh_RGB And the proportion of white pixels to the entire image size , retain a suitable difference binary image img_new for subsequent analysis and processing. The specific algorithm selection logic is as follows:

If a1=0 and b2>0.5, select the channel difference binary image diff_RGB as the difference binary image img_new;

If a2=0 and b1>0.5, select the channel maximum value average difference binary image thresh_RGB as the difference binary image img_new;

If a1=0 and a2>0, select the channel maximum value average difference binary image thresh_RGB as the difference binary image img_new;

If a2=0 and a1>0, select the channel difference binary image diff_RGB as the difference binary image img_new.

S3: Image change area analysis. Since the device occupies a small area in the point map, more than half of the area in the picture is background, but the change of natural background is not considered as device change, and no alarm is required. Therefore, the selected difference binary image img_new needs to be analyzed for image change areas, including corrosion, seasonal background change elimination, expansion, light and shadow elimination, connected domain filtering, and change frame merging and restoration.

Among them, seasonal background change elimination is used to exclude the presence of many point-like change connected domains in the difference binary map img_new due to seasonal background changes. For example, the grass around the device changes from yellow to green due to seasonal changes, and there is snow on the ground in winter. The densely distributed connected domains in the difference binary map img_new interfere with the device change analysis. The specific steps include:

S31: Find all the point-connected domains in the difference binary image img_new, count the number of point-connected domains, if the number of point-connected domains is greater than 30, and the area of each connected domain is less than 200 pixels, fill these point-connected domains as background, count them as unchanged areas, to exclude interference from non-device point-connected domains, and obtain the difference binary image img_new1; perform image erosion processing on the difference binary image img_new1, and obtain the difference binary image img_erode. Erosion processing is a well-known technology in the art and will not be described in detail here.

Light shadow culling is used to eliminate the situation where the shadow position in the difference binary image img_new1 is different due to light changes, but the device itself has not changed. For example, the position of the device shadow changes due to the change in the position of the sun in the morning and afternoon, which leads to more horizontal and vertical elongated connected domains in the difference binary image img_new1. The specific steps include the following:

S32: Count the number of connected domains in the difference binary map img_erode and the difference binary map img_new1, and calculate the ratio of the total number of pixels of all connected domains in the two maps to the total circumscribed rectangular frame area of all connected domains. If the ratio is less than 0.4, and the number of connected domains in the difference binary map img_new1 is greater than 20 and the number of connected domains in the difference binary map img_erode is less than 3, it proves that the number of pixels in the change frame is too small and mainly thin and long lines. Then fill the connected domains in the img_erode map as the background and count them as unchanged connected domains to obtain the difference binary map img_new2; perform image dilation processing on the difference binary map img_new2 to obtain the difference binary map img_dilate. Dilation processing is a well-known technology in the art and will not be repeated here.

Connected domain filtering does not analyze and process a certain type of interference, but is used to exclude all interference that does not match the shape characteristics of the substation equipment. It specifically includes the following steps:

S33: Fill the connected domains with an area less than 200 pixels and a duty ratio less than 0.2 and the line segment connected domains with an aspect ratio greater than 5 as the background, and count them as unchanged, to obtain the difference binary image img_new3. The duty ratio less than 0.2 means that the ratio of the number of pixels to the external rectangular frame is less than 0.2, and the aspect ratio greater than 5 means: width/height>5 or height/width>5, because there is no common equipment monitored in the substation field with an aspect ratio too large.

The change frame merging and restoration specifically includes the following steps:

S34: merge the external rectangular frames corresponding to all the remaining connected domains of the difference binary image img_new3 into an output change frame, take the four maximum values of the coordinates of all the external rectangular frames: x _min , x _max , y _min , y _max , and obtain the output change frame coordinates ((x _min , y _min ), (x _max , y _max )) to obtain the output change frame; expand the output change frame by 4 times to 1920 The changing box coordinates in 1080 pixels.

S4: Frame the changed area on the point map where the change occurs and issue an alarm.

The method in the technical solution of the present invention was used for testing, and the test results are shown in Table 1:

Table 1: Accuracy test results of the method of the present invention in different scenarios

The above illustrations prove that the present invention is sensitive to device change areas and can eliminate interference from natural background changes. Abnormal change areas can be identified on 1080-pixel images with high accuracy. In addition, the experiments were conducted using a variety of different device scenarios, which also proves that the present invention has good versatility and does not require long-term model training when adding new scenarios.

In order to compare with the experimental results of the present invention, a comparative test was also conducted, and the measurement indicators are shown in Table 2:

Table 2: Comparative test results of different methods

Through comparative experiments, it can be seen that the change detection algorithm based on deep learning has a slight advantage in accuracy due to the model obtained after a long period of training, but has a significant disadvantage in speed compared with the traditional algorithm. The algorithm adopted in the present invention has a relatively good performance in terms of comprehensive accuracy and speed.

The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. The intelligent analysis method for the bitmap change area of the substation based on image comparison comprises the steps of determining a plurality of patrol points in a patrol area, coding each point, and setting a shooting device at the patrol points; establishing a reference drawing library of each point position in the inspection area; when the patrol task is executed, a shooting device is utilized to capture a point map of patrol points; extracting a reference image in a reference image library according to point position codes and comparing the extracted reference image with a corresponding point bitmap; the method is characterized by comprising the following steps of:

S1: image preprocessing, wherein the image preprocessing comprises image downsampling, key point screening and point map offset correction;

S2: screening according to color characteristics, wherein the screening according to the color characteristics comprises RGB color channel difference making, RGB color channel maximum value difference making with average value difference making and a difference value diagram selecting algorithm;

S3: the method comprises the steps of analyzing an image change area, wherein the image change area comprises corrosion, seasonal background change elimination, expansion, light shadow elimination, connected domain filtration and change frame merging and reduction;

s4: and framing a change area on the changed point bitmap, and alarming.

2. The intelligent analysis method for the change area of the bitmap of the substation based on the image comparison according to claim 1, wherein the image downsampling comprises the steps of compressing the aspect ratio of the reference map to 1/4 of the original aspect ratio to obtain the reference map img1; the image downsampling further comprises compressing the aspect ratio of the bitmap to 1/4 of the original to obtain a bitmap img2.

3. The intelligent analysis method for the bitmap change region of the transformer station based on image comparison according to claim 2, wherein the key point screening step is as follows:

S11: performing feature point detection on the reference graph img1 and the point graph img2 by using a feature point matching algorithm based on sift, and calculating a feature descriptor for each feature point;

s12: identifying the nearest neighbor relation between the feature descriptors of the reference graph img1 and the point graph img2 by a violent matching method, and introducing a K nearest neighbor matching strategy to determine the best matching feature points and the suboptimal matching feature points in order to ensure the matching accuracy;

s13: and setting a distance ratio filtering threshold value, and completing screening of the key point pairs by utilizing the distance ratio between the best matching characteristic point and the next best matching characteristic point and the corresponding characteristic point.

4. The intelligent analysis method for the bitmap change region of the substation based on image comparison according to claim 2, wherein the point map offset correction comprises:

If the number of the screened key point pairs exceeds 4, calculating an optimal single mapping transformation matrix between the key point pairs, and carrying out reduction alignment in X and Y directions on the point bitmap according to the optimal single mapping transformation matrix to generate a registration graph img_s which is spatially aligned with the reference graph img 1; if the number of the reserved characteristic point pairs is less than 4, the point map img2 is directly used as a registration map img_s.

5. The intelligent analysis method for the bitmap change region of the transformer station based on the image comparison according to claim 4, wherein the filtering according to the color characteristics comprises the following steps:

S21: directly differencing the reference image img1 and the registration image img_s to obtain a difference image img_diff;

S22: the RGB color channel difference comprises separating R, G, B channels of a difference image img_diff, calculating absolute difference values diff_R_ G, diff _G_ B, diff _B_R between every two red-green, green-blue and blue-red channels, and creating a mask for each difference value; setting a threshold range one as 35 to 255, reserving pixel values in the threshold range one in masks, and combining 3 masks to obtain: channel difference binary pattern diff_rgb;

S23: the difference between the maximum value and the average value of the RGB color channels comprises: separating R, G, B channels of the difference image img_diff, taking average values R_mean, G_mean and B_mean of 3 channels, and calculating absolute differences R-R_mean, G-G_mean and B-B_mean of R, G, B and the average values of the channels respectively; and then the maximum values R_max, G_max and B_max of R, G, B channels are respectively taken, and the following formulas are adopted: Obtaining a lowest threshold value rgb_scale reserved by the pixel point; finally, creating masks for R-R_mean, G-G_mean and B-B_mean of each channel, setting a threshold range II as rgb_scale to 255, reserving pixels in the threshold range II in the masks, and merging 3 masks to obtain the pixel structure: channel maximum average difference binary image thresh_rgb.

6. The intelligent analysis method for the change area of the substation bitmap based on the image comparison according to claim 5, wherein a difference value graph selection algorithm is designed according to the color characteristics of equipment in a scene image, a difference value binary graph img_new is selected, the difference value graph selection algorithm comprises the steps of calculating the numbers a1 and a2 of white pixels and the proportions b1 and b2 of the white pixels in the channel difference value binary graph diff_RGB and the channel maximum value average value difference value binary graph thresh_RGB, and the specific algorithm selection logic is as follows:

if a1=0 and b2>0.5, diff_rgb is selected as the difference binary image img_new;

If a2=0 and b 1> 0.5, selecting thresh_rgb as a difference binary image img_new;

if a1=0 and a2 >0, selecting thresh_rgb as a difference binary image img_new;

If a2=0 and a1>0, diff_rgb is selected as the difference binary image img_new.

7. The intelligent analysis method for the bitmap change region of the substation based on image comparison according to claim 6, wherein the removing of the seasonal background change comprises: finding all punctiform connected domains in the difference binary image img_new, counting the number of punctiform connected domains, and filling the punctiform connected domains into a background to obtain the difference binary image img_new1 if the number of punctiform connected domains is more than 30 and the area of each connected domain is smaller than 200 pixel points; and performing image corrosion processing on the difference binary image img_new1 to obtain a difference binary image img_ erode.

8. The intelligent analysis method for the bitmap change region of the substation based on image comparison according to claim 7, wherein the light shadow elimination comprises counting the number of connected domains in a difference binary image img_ erode and a difference binary image img_new1, respectively calculating the ratio of the total pixel number of all the connected domains in the two images to the total external rectangular frame area of all the connected domains, and if the ratio is smaller than 0.4, the number of the connected domains in the difference binary image img_new1 is larger than 20, and the number of the connected domains in the difference binary image img_ erode is smaller than 3, filling the connected domains in the img_ erode image as a background to obtain a difference binary image img_new2; and performing image expansion processing on the difference binary image img_new2 to obtain a difference binary image img_ dilate.

9. The intelligent analysis method for the bitmap change region of the transformer station based on image comparison according to claim 8, wherein the connected domain filtering comprises filling connected domains with an area smaller than 200 pixels, a duty ratio smaller than 0.2 and a line segment connected domain with an aspect ratio larger than 5 as a background to obtain a difference binary image img_new3.

10. The intelligent analysis method for the bitmap change region of the substation based on image comparison according to claim 9, wherein the merging and restoring of the change frames comprises merging all the external rectangular frames corresponding to all the remaining connected domains of the difference binary image img_new3 into one change frame, and taking 4 maximum values of coordinates of all the external rectangular frames: x _min、x_max、y_min、y_max, obtaining an output change frame coordinate ((x _min,y_min）,（x_max,y_max)); enlarging the output change frame by 4 times to 19201080 Pixels.