CN115082745A

CN115082745A - Image-based cable strand quality detection method and system

Info

Publication number: CN115082745A
Application number: CN202211006052.0A
Authority: CN
Inventors: 尹灵; 邓旺; 杨英合; 彭茂华; 杨东生; 施传新
Original assignee: SHENZHEN CHENGTIANTAI CABLE INDUSTRIAL DEVELOPMENT CO LTD
Current assignee: SHENZHEN CHENGTIANTAI CABLE INDUSTRIAL DEVELOPMENT CO LTD
Priority date: 2022-08-22
Filing date: 2022-08-22
Publication date: 2022-09-20
Anticipated expiration: 2042-08-22
Also published as: CN115082745B

Abstract

Disclosed are a method and a system for detecting the quality of a cable strand based on an image, which realize the intellectualization of the quality detection of the cable strand. Specifically, a plurality of stranded wire section images of a cable stranded wire to be detected are collected along the extending direction of the cable stranded wire to be detected through a camera, then the characteristics of each stranded wire section are extracted through a first convolution neural network with a multi-scale convolution structure through the stranded wire section images, the difference between the characteristics of different stranded wire sections is obtained through covariance calculation, then a second convolution neural network with a three-dimensional convolution kernel is used as a characteristic extractor to capture the correlation characteristics between the differences of the characteristics of the different stranded wire sections so as to obtain a stranded wire section correlation characteristic diagram, and finally a classification result used for indicating whether the cable stranded wire to be detected has quality defects is obtained through a classifier on the stranded wire section correlation characteristic diagram. Therefore, an intelligent detection scheme for the quality of the cable stranded wire is constructed based on the stranded wire image.

Description

Image-based cable strand quality detection method and system

Technical Field

The present application relates to the field of intelligent manufacturing, and more particularly, to a method and system for detecting quality of a cable strand based on an image.

Background

The cable stranded wire is a conductive inner core of the wire cable, and commonly used copper wires and aluminum wires are used, and the copper wires and the aluminum wires can be stranded into wire cores of various wire cables with different specifications and sections. During the preparation of cable strands, various types of defects occur due to various reasons, for example, 1. surface scratches of single or stranded wires; 2. single line peeling, scar, brittle fracture, arching and inclusion; 3. the twisting direction is wrong, the shape is snake-shaped, the twisting pitch is large, and the length is unqualified; and 4, arranging the wires in disorder, crushing, scratching, collision, unqualified direct current resistance of the conductive wire core of the wire and the cable, and the like.

The traditional cable strand quality detection is mainly realized by means of human eye observation, the method is very dependent on the experience of workers, and the method is limited by the observation resolution of human eyes, so that fine or hidden defects cannot be observed.

Therefore, an optimized quality detection scheme for the cable strands is desired.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a method and a system for detecting the quality of a cable strand based on an image, which realize the intellectualization of the quality detection of the cable strand. Specifically, a plurality of stranded wire section images of a cable stranded wire to be detected are collected along the extending direction of the cable stranded wire to be detected through a camera, then the characteristics of each stranded wire section are extracted through a first convolution neural network with a multi-scale convolution structure through the stranded wire section images, the difference between the characteristics of different stranded wire sections is obtained through covariance calculation, then a second convolution neural network with a three-dimensional convolution kernel is used as a characteristic extractor to capture the correlation characteristics between the differences of the characteristics of the different stranded wire sections so as to obtain a stranded wire section correlation characteristic diagram, and finally a classification result used for indicating whether the cable stranded wire to be detected has quality defects is obtained through a classifier on the stranded wire section correlation characteristic diagram. Therefore, an intelligent detection scheme for the quality of the cable stranded wire is constructed based on the stranded wire image.

According to an aspect of the present application, there is provided an image-based cable strand quality detection method, including: collecting a plurality of stranded wire section images of a cable stranded wire to be detected along the extending direction of the cable stranded wire to be detected; respectively passing each stranded wire section image in a plurality of stranded wire section images of the cable stranded wire to be detected through a first convolution neural network with a multi-scale convolution structure to obtain a plurality of multi-scale stranded wire section characteristic graphs; performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; calculating a covariance matrix between every two multiscale stranded wire section eigenvectors in the multiscale stranded wire section eigenvectors to obtain a plurality of covariance matrices; correcting the eigenvalue of each position in each covariance matrix in the plurality of covariance matrices to obtain a plurality of corrected covariance matrices; arranging the corrected covariance matrixes into a three-dimensional input tensor, and then obtaining a twisted line section association characteristic diagram through a second convolution neural network with a three-dimensional convolution kernel; and the stranded wire section association characteristic graph is processed by a classifier to obtain a classification result, and the classification result is used for indicating whether the cable stranded wire to be detected has quality defects or not.

In the image-based method for detecting the quality of the cable strand, the step of passing each strand segment image of a plurality of strand segment images of the cable strand to be detected through a first convolution neural network with a multi-scale convolution structure to obtain a plurality of multi-scale strand segment characteristic diagrams includes: respectively performing the following on input data in the forward transfer process of the layers by using each layer of the first convolutional neural network with the multi-scale convolutional structure: performing convolution processing based on a first convolution kernel on the input data to obtain a first scale convolution characteristic diagram; performing convolution processing based on a second convolution kernel on the input data to obtain a second scale convolution characteristic diagram; performing convolution processing on the input data based on a third convolution kernel to obtain a third scale convolution characteristic diagram; cascading the first scale convolution feature map, the second scale convolution feature map and the third scale convolution feature map to obtain a multi-scale convolution feature map; pooling the multi-scale convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooled feature map to obtain a nonlinear activation feature map; and outputting the final layer of the first convolutional neural network with the multi-scale convolutional structure as the multi-scale twisted line segment characteristic diagram.

In the above method for detecting quality of a cable strand based on an image, the performing global mean pooling along a channel dimension on each of the multiple multi-scale strand segment feature maps to obtain multiple multi-scale strand segment feature vectors includes: and performing global mean pooling on each characteristic matrix along the channel dimension of each multi-scale stranded wire section characteristic diagram in the multi-scale stranded wire section characteristic diagrams to obtain the multi-scale stranded wire section characteristic vectors.

In the above method for detecting quality of a twisted cable based on an image, the correcting the eigenvalue of each position in each covariance matrix of the covariance matrices to obtain a plurality of corrected covariance matrices includes: correcting the eigenvalue of each position in each covariance matrix in the covariance matrices according to the following formula to obtain a plurality of corrected covariance matrices; wherein the formula is:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is said plurality of covariance matrices divided by said second

Eigenvalues of corresponding positions of other covariance matrices than the individual covariance matrix,

the hyper-parameters are controlled for the space.

In the image-based method for detecting the quality of a twisted cable, the arranging the corrected covariance matrices into a three-dimensional input tensor, and then obtaining a twisted cable segment associated feature map through a second convolutional neural network with a three-dimensional convolutional kernel includes: performing convolution processing, pooling processing and nonlinear activation processing on input data in forward pass of layers respectively by using the layers of the second convolutional neural network with the three-dimensional convolutional kernel, so as to output the twisted wire segment correlation feature map by the last layer of the second convolutional neural network with the three-dimensional convolutional kernel, wherein the input of the first layer of the second convolutional neural network with the three-dimensional convolutional kernel is the three-dimensional input tensor.

In the above method for detecting quality of a cable strand based on an image, the passing the strand segment association feature map through a classifier to obtain a classification result includes: processing the twisted wire segment association feature map by using the classifier according to the following formula to obtain the classification result, wherein the formula is as follows:

wherein the content of the first and second substances,

in order to be a result of said classification,

and

is as follows

The weights and bias matrices corresponding to the individual classes,

and associating a characteristic diagram with the stranded wire section.

According to another aspect of the present application, there is provided an image-based cable strand quality detection system, comprising: the image acquisition module is used for acquiring a plurality of stranded wire section images of the cable stranded wire to be detected along the extending direction of the cable stranded wire to be detected; the first characteristic extraction module is used for enabling each stranded wire section image in the stranded wire section images of the cable stranded wire to be detected to pass through a first convolution neural network with a multi-scale convolution structure respectively to obtain a plurality of multi-scale stranded wire section characteristic graphs; the pooling processing module is used for performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; the covariance matrix calculation module is used for calculating a covariance matrix between every two multiscale stranded wire section eigenvectors in the multiscale stranded wire section eigenvectors to obtain a plurality of covariance matrices; the correction module is used for correcting the eigenvalue of each position in each covariance matrix in the covariance matrices to obtain a plurality of corrected covariance matrices; the second feature extraction module is used for arranging the corrected covariance matrixes into a three-dimensional input tensor and then obtaining a stranded wire section association feature map through a second convolution neural network with a three-dimensional convolution kernel; and the detection result generation module is used for enabling the stranded wire section correlation characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the cable stranded wire to be detected has quality defects or not.

In the image-based cable strand quality detection system, the first feature extraction module is configured to perform, on the input data, in a layer forward transfer process using each layer of the first convolutional neural network having the multi-scale convolutional structure: performing convolution processing based on a first convolution kernel on the input data to obtain a first scale convolution characteristic diagram; performing convolution processing based on a second convolution kernel on the input data to obtain a second scale convolution characteristic diagram; performing convolution processing based on a third convolution kernel on the input data to obtain a third scale convolution characteristic diagram; cascading the first scale convolution feature map, the second scale convolution feature map and the third scale convolution feature map to obtain a multi-scale convolution feature map; pooling the multi-scale convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooling characteristic map to obtain a nonlinear activation characteristic map; and outputting the final layer of the first convolutional neural network with the multi-scale convolutional structure as the multi-scale twisted line segment characteristic diagram.

In the image-based cable strand quality detection system, the pooling processing module is configured to perform global mean pooling processing on each feature matrix along a channel dimension of each multi-scale strand segment feature map in the multiple multi-scale strand segment feature maps respectively to obtain the multiple multi-scale strand segment feature vectors.

In the image-based cable strand quality detection system, the correction module is configured to correct feature values of each position in each of the plurality of covariance matrices according to the following formula to obtain a plurality of corrected covariance matrices; wherein the formula is:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is said plurality of covariance matrices divided by said second

the hyper-parameters are controlled for the space.

In the above image-based cable strand quality detection system, the second feature extraction module is configured to perform convolution processing, pooling processing and nonlinear activation processing on input data in forward pass of layers respectively using layers of the second convolutional neural network with the three-dimensional convolutional kernel to output the strand segment associated feature map from a last layer of the second convolutional neural network with the three-dimensional convolutional kernel, where an input of the first layer of the second convolutional neural network with the three-dimensional convolutional kernel is the three-dimensional input tensor.

In the image-based cable strand quality detection system, the detection result generation module is configured to process the strand segment association feature map by using the classifier according to the following formula to obtain the classification result, where the formula is:

wherein, the first and the second end of the pipe are connected with each other,

in order to be the result of said classification,

and

is a first

The weights and bias matrices corresponding to the individual classes,

and associating a characteristic diagram with the stranded wire section.

According to still another aspect of the present application, there is provided an electronic apparatus including: a processor; and a memory having stored therein computer program instructions which, when executed by the processor, cause the processor to execute the image-based cable strand quality detection method as described above.

According to yet another aspect of the present application, there is provided a computer readable medium having stored thereon computer program instructions, which, when executed by a processor, cause the processor to execute the image-based cable strand quality detection method as described above.

Compared with the prior art, the image-based cable strand quality detection method and the image-based cable strand quality detection system have the advantage that the intellectualization of the quality detection of the cable strands is realized. Specifically, a plurality of stranded wire section images of a cable stranded wire to be detected are collected along the extending direction of the cable stranded wire to be detected through a camera, then the characteristics of each stranded wire section are extracted through a first convolution neural network with a multi-scale convolution structure through the stranded wire section images, the difference between the characteristics of different stranded wire sections is obtained through covariance calculation, then a second convolution neural network with a three-dimensional convolution kernel is used as a characteristic extractor to capture the correlation characteristics between the differences of the characteristics of the different stranded wire sections so as to obtain a stranded wire section correlation characteristic diagram, and finally a classification result used for indicating whether the cable stranded wire to be detected has quality defects is obtained through a classifier on the stranded wire section correlation characteristic diagram. Therefore, an intelligent detection scheme for the quality of the cable stranded wire is constructed based on the stranded wire image.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.

Fig. 1 is an application scenario diagram of a cable strand quality detection method based on an image according to an embodiment of the present application.

Fig. 2 is a flowchart of an image-based cable strand quality detection method according to an embodiment of the present application.

Fig. 3 is a schematic configuration diagram of a method for detecting quality of a twisted cable based on an image according to an embodiment of the present application.

Fig. 4 is a flowchart of obtaining a plurality of multi-scale stranded wire segment characteristic diagrams by respectively passing each stranded wire segment image in a plurality of stranded wire segment images of the cable stranded wire to be detected through a first convolution neural network having a multi-scale convolution structure in the image-based cable stranded wire quality detection method according to the embodiment of the application.

Fig. 5 is a block diagram of an image-based cable strand quality detection system according to an embodiment of the present application.

Fig. 6 is a block diagram of an electronic device according to an embodiment of the application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.

Overview of scenes

As described above, the conventional quality inspection of the cable strands is mainly realized by means of human-eye observation, which is very dependent on the experience of workers and is limited by the observation resolution of human eyes, so that fine or hidden defects cannot be observed. Therefore, an optimized quality detection scheme for the cable strands is desired.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.

Deep learning and development of a neural network provide a new solution for quality detection of the cable strands.

It should be understood that the quality detection of the cable strands can be converted into a multi-classification problem based on images, that is, a deep neural network is used as a feature extractor to extract features in the images of the cable strands to be detected, and the extracted features are expressed through a classifier to obtain a classification result for expressing whether the cable strands to be detected have quality defects or not. However, considering that the defects of the cable strands are numerous and the features of the cable strands are not obvious in the image level, it is difficult to improve the accuracy of quality detection if the model is directly combined with a simple feature extractor and classifier.

Therefore, in the technical scheme of the application, the correlation mode between the comparison between the characteristics of different stranded wire sections and the characteristic difference is used as the basis for classification judgment. It should be understood that if a certain twisted line segment has a quality defect and other twisted line segments do not have a quality defect, the contrast between the two will be obvious, and if the difference characteristics are close, it indicates that different twisted line segments may have the same defect type.

Specifically, firstly, a plurality of stranded wire section images of the cable stranded wire to be detected are collected along the extending direction of the cable stranded wire to be detected through a camera. Then, a convolutional neural network model is used as a feature extractor to extract features of the twisted line segment images respectively. In order to accurately extract the characteristics of the defect area, the convolutional neural network model used is structurally improved. In particular, a multi-scale convolution structure is introduced.

The multi-scale convolution structure exists in each layer of the convolution neural network model and comprises a plurality of convolution kernels with different sizes, and the convolution kernels with different receptive fields respectively extract the features of the input data, so that the convolution kernel with a large receptive field can extract more context information, and the convolution kernel with a small receptive field can extract features with smaller granularity, and the perception capability of defect features can be improved by combining the context information and the granularity features.

And then, performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors. Namely, the multi-scale stranded wire segment feature map is subjected to dimensionality reduction treatment in a global mean pooling mode.

Then, a covariance matrix between every two multiscale stranded wire section eigenvectors in the multiscale stranded wire section eigenvectors is calculated to obtain a plurality of covariance matrices, that is, a covariance matrix table between every two multiscale stranded wire section eigenvectors is used for representing the differential expression between the high-dimensional implicit features between the two stranded wire sections. And arranging the plurality of covariance matrixes into a three-dimensional input tensor, and then obtaining a twisted line section association feature map through a second convolution neural network with a three-dimensional convolution kernel, namely capturing association features among twisted line sections of differences among local features of high-dimensional images of different twisted line sections by using the second convolution neural network with the three-dimensional convolution kernel as a feature extractor. And then, obtaining a classification result for indicating whether the cable strand to be detected has quality defects or not by using the strand segment correlation characteristic diagram through a classifier.

Here, for the plurality of covariance matrices, since each covariance matrix is a covariance matrix between feature vectors of two multi-scale twisted wire segments, which is a covariance representation of distributed image multi-scale semantics along the extending direction of the cable twisted wire to be detected, there is a certain degree of anisotropy in feature distribution, that is, it represents a narrow subset residing in the entire high-dimensional feature space, which makes the arrangement lack continuity after three-dimensional input tensor, affecting the feature extraction effect of the second convolutional neural network using a three-dimensional convolution kernel in the channel dimension.

Therefore, each covariance matrix is first aligned with the search space, i.e.:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is said plurality of covariance matrices divided by said second

And the eigenvalues of the corresponding positions of other covariance matrixes except the individual covariance matrix.

For spatial control of hyper-parameters, e.g. initially set to

The individual covariance matrix and

the distance between the individual covariance matrices.

By comparing search space syntropy, each covariance matrix can be transferred to an isotropic and differentiated expression space so as to enhance distribution continuity of feature expression of the three-dimensional input tensor obtained after arrangement along the channel direction and improve feature expression effect of the twisted-wire section associated feature map. Thus, the accuracy of quality detection of the cable strand is improved.

Based on this, the application provides an image-based cable strand quality detection method, which includes: collecting a plurality of stranded wire section images of a cable stranded wire to be detected along the extending direction of the cable stranded wire to be detected; enabling each stranded wire section image in the plurality of stranded wire section images of the cable stranded wire to be detected to pass through a first convolution neural network with a multi-scale convolution structure to obtain a plurality of multi-scale stranded wire section characteristic diagrams; performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; calculating a covariance matrix between every two multiscale stranded wire section eigenvectors in the multiscale stranded wire section eigenvectors to obtain a plurality of covariance matrices; correcting the eigenvalue of each position in each covariance matrix in the plurality of covariance matrices to obtain a plurality of corrected covariance matrices; arranging the corrected covariance matrixes into a three-dimensional input tensor, and then obtaining a stranded wire section association feature map through a second convolution neural network with a three-dimensional convolution kernel; and the stranded wire section association characteristic graph is processed by a classifier to obtain a classification result, and the classification result is used for indicating whether the cable stranded wire to be detected has quality defects or not.

Fig. 1 is an application scenario diagram of a cable strand quality detection method based on an image according to an embodiment of the present application. As shown in fig. 1, in this application scenario, a plurality of stranded wire segment images of a cable stranded wire to be detected (e.g., M as illustrated in fig. 1) are first acquired by a camera (e.g., C as illustrated in fig. 1) along an extending direction of the cable stranded wire to be detected; then, the multiple stranded wire segment images of the cable stranded wire to be detected are input into a server (for example, S shown in fig. 1) deployed with an image-based cable stranded wire quality detection algorithm, wherein the server processes the multiple stranded wire segment images of the cable stranded wire to be detected with the image-based cable stranded wire quality detection algorithm to output a classification result, and the classification result is used for indicating whether the cable stranded wire to be detected has a quality defect.

Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary method

Fig. 2 is a flowchart of an image-based cable strand quality detection method according to an embodiment of the present application. As shown in fig. 2, the method for detecting quality of a twisted cable based on an image according to an embodiment of the present application includes: s110, collecting a plurality of stranded wire section images of a cable stranded wire to be detected along the extending direction of the cable stranded wire to be detected; s120, enabling each stranded wire segment image in the stranded wire segment images of the cable stranded wire to be detected to pass through a first convolution neural network with a multi-scale convolution structure to obtain a plurality of multi-scale stranded wire segment characteristic graphs; s130, performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; s140, calculating a covariance matrix between every two multiscale stranded wire segment eigenvectors in the multiscale stranded wire segment eigenvectors to obtain a plurality of covariance matrices; s150, correcting the eigenvalue of each position in each covariance matrix in the covariance matrices to obtain a plurality of corrected covariance matrices; s160, arranging the corrected covariance matrixes into a three-dimensional input tensor, and then obtaining a stranded wire section association feature map through a second convolution neural network with a three-dimensional convolution kernel; and S170, passing the twisted wire section correlation characteristic diagram through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the cable twisted wire to be detected has quality defects or not.

Fig. 3 is a schematic configuration diagram of a method for detecting quality of a twisted cable based on an image according to an embodiment of the present application. As shown in fig. 3, in the framework of the image-based method for detecting the quality of a cable strand, first, a plurality of strand segment images of the cable strand to be detected are collected along the extending direction of the cable strand to be detected; then, enabling each stranded wire segment image in the plurality of stranded wire segment images of the cable stranded wire to be detected to pass through a first convolution neural network with a multi-scale convolution structure respectively to obtain a plurality of multi-scale stranded wire segment characteristic graphs; then, performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; then, calculating a covariance matrix between every two multi-scale stranded wire section eigenvectors in the multi-scale stranded wire section eigenvectors to obtain a plurality of covariance matrices; then, correcting the eigenvalue of each position in each covariance matrix in the plurality of covariance matrices to obtain a plurality of corrected covariance matrices; then, arranging the corrected covariance matrixes into a three-dimensional input tensor, and then obtaining a stranded wire section association feature map through a second convolution neural network with a three-dimensional convolution kernel; and finally, passing the twisted wire section association characteristic diagram through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the cable twisted wire to be detected has quality defects or not.

In step S110, a plurality of stranded wire segment images of the cable stranded wire to be detected are collected along the extending direction of the cable stranded wire to be detected. As mentioned above, it is considered that the quality detection of the cable strand by human eye is limited by the observation resolution of human eye, so that the fine or hidden defects on the cable strand cannot be observed. In the technical scheme of the application, the quality detection of the cable strands is converted into the problem of multi-classification based on images, namely, the deep neural network is used as a feature extractor to extract features in the images of the cable strands to be detected, and the extracted features are expressed to obtain a classification result for expressing whether the cable strands to be detected have quality defects through a classifier.

However, considering that the defects of the cable strands are numerous and the features of the cable strands are not obvious in the image-level presentation, it is difficult to improve the accuracy of quality detection if the model is directly combined with a simple feature extractor and classifier. Therefore, in the technical scheme of the application, the correlation mode between the comparison between the characteristics of different stranded wire sections and the characteristic difference is used as the basis for classification judgment. It should be understood that if a certain twisted line segment has a quality defect and other twisted line segments do not have a quality defect, the contrast between the two will be obvious, and if the difference characteristics are close, it indicates that different twisted line segments may have the same defect type.

It can be understood that, in the embodiment of the present application, a plurality of stranded conductor segment images of a cable stranded conductor to be detected are acquired by a camera along an extending direction of the cable stranded conductor to be detected. The number of the cameras is not limited to one illustrated in fig. 1, that is, the number of the cameras may be multiple, and multiple cameras may be arranged along the extending direction of the cable strand to be detected so as to collect multiple strand section images of the cable strand to be detected respectively, so as to improve the image collection efficiency.

In step S120, each stranded conductor segment image in the plurality of stranded conductor segment images of the cable stranded conductor to be detected is respectively passed through a first convolution neural network having a multi-scale convolution structure to obtain a plurality of multi-scale stranded conductor segment characteristic diagrams. That is, the first convolutional neural network model is used as a feature extractor to perform feature extraction on each twisted line segment image in the plurality of twisted line segment images. It can be understood that, in the technical solution of the present application, in order to accurately extract the features of the region where the defect is located, the first convolution neural network model used is structurally improved. For example, in a specific example of the present application, a multi-scale convolution structure is introduced.

The multi-scale convolution structure exists in each layer of the first convolution neural network model and comprises a plurality of convolution kernels with different sizes, and the convolution kernels with different receptive fields respectively extract the features of the input data, so that the convolution kernel with a large receptive field can extract more context information, and the convolution kernel with a small receptive field can extract features with smaller granularity, and the perception capability of defect features can be improved by combining the context information and the granularity features.

Specifically, in this embodiment of the present application, fig. 4 is a flowchart illustrating that, in the method for detecting quality of a cable strand based on an image according to the embodiment of the present application, each strand segment image in a plurality of strand segment images of the cable strand to be detected is respectively passed through a first convolutional neural network having a multi-scale convolutional structure to obtain a plurality of multi-scale strand segment characteristic diagrams. As shown in fig. 4, the obtaining of a plurality of multi-scale stranded wire segment characteristic diagrams by respectively passing each stranded wire segment image in a plurality of stranded wire segment images of the cable stranded wire to be detected through a first convolution neural network having a multi-scale convolution structure includes: respectively performing the following steps on input data in the forward transfer process of the layers by using each layer of the first convolutional neural network with the multi-scale convolution structure: s210, performing convolution processing on the input data based on a first convolution kernel to obtain a first scale convolution characteristic diagram; s220, performing convolution processing on the input data based on a second convolution kernel to obtain a second scale convolution characteristic diagram; s230, performing convolution processing on the input data based on a third convolution kernel to obtain a third scale convolution characteristic diagram; s240, cascading the first scale convolution feature map, the second scale convolution feature map and the third scale convolution feature map to obtain a multi-scale convolution feature map; s250, performing pooling processing on the multi-scale convolution characteristic graph to obtain a pooled characteristic graph; s260, carrying out nonlinear activation on the pooling characteristic map to obtain a nonlinear activation characteristic map; and outputting the final layer of the first convolutional neural network with the multi-scale convolutional structure as the multi-scale twisted line segment characteristic diagram.

Further, in order to accurately extract the characteristics of each stranded wire segment image in the plurality of stranded wire segment images of the cable stranded wire to be detected, a small-sized convolution kernel is provided in the application to solve the problem. In addition, besides the characteristics of the cable strand to be detected, abundant context information exists around the cable strand to be detected, for example, some characteristics specific to an outer protection layer of the cable strand. The characteristic information can effectively supplement the characteristics of the cable strands and help to better confirm whether the cable strands to be detected have quality defects. That is, in the present application, in order to extract contextual feature information around the cable strand to be detected, convolution kernels of different sizes are required to be used to extract features on the images of the respective strand segments.

It can be understood that the multi-scale convolution kernel effectively solves the problem that a single-size convolution kernel cannot extract features of different scales in an image, but each size convolution kernel still extracts features only once in the convolution layer. In order to achieve the purpose of extracting features for multiple times, a concept of grouping convolution is proposed: and respectively extracting features from the feature maps by utilizing the grouping convolution, dividing a convolution kernel into a plurality of groups according to channels by utilizing the grouping convolution, and respectively carrying out convolution operation on the feature maps. In another specific example of the present application, one convolution kernel is divided into multiple groups of convolution kernels according to channels, and multiple groups of convolution kernels based on different channels are respectively performed on each stranded wire segment image in the multiple stranded wire segment images to obtain the multiple multi-scale stranded wire segment characteristic diagrams.

In step S130, global mean pooling along the channel dimension is performed on each multi-scale stranded wire segment feature map in the multiple multi-scale stranded wire segment feature maps, respectively, to obtain multiple multi-scale stranded wire segment feature vectors. Namely, the multi-scale stranded wire segment feature map is subjected to dimensionality reduction treatment in a global mean pooling mode.

Specifically, in the embodiment of the present application, global mean pooling is performed on each feature matrix along the channel dimension of each multi-scale stranded wire segment feature map in the multiple multi-scale stranded wire segment feature maps, so as to obtain the multiple multi-scale stranded wire segment feature vectors. Here, the multiple multi-scale stranded wire segment feature vectors are obtained through global mean calculation, so that the global feature information of the multiple multi-scale stranded wire segment feature maps can be focused on the basis of dimensionality reduction to prevent the reduction of classification accuracy caused by data loss, and the number of parameters can be reduced, so that the calculated amount is reduced, overfitting is prevented, and the accuracy of subsequent classification is improved.

In steps S140 and S150, a covariance matrix between every two multi-scale twisted-line segment eigenvectors in the multi-scale twisted-line segment eigenvectors is calculated to obtain a plurality of covariance matrices, and then eigenvalues of each position in each covariance matrix in the plurality of covariance matrices are corrected to obtain a plurality of corrected covariance matrices. That is, the difference expression between the high-dimensional implicit features between two twisted line segments is represented by a covariance matrix table between every two multi-scale twisted line segment feature vectors.

Specifically, for the plurality of covariance matrices, each covariance matrix is a covariance matrix between feature vectors of two multi-scale twisted wire segments, and therefore, as a covariance representation of distributed image multi-scale semantics along the extending direction of the cable twisted wire to be detected, there is a certain degree of anisotropy in feature distribution, that is, it represents a narrow subset residing in the whole high-dimensional feature space, which makes the arrangement lack continuity after three-dimensional input tensor, and affects the feature extraction effect of the second convolutional neural network using a three-dimensional convolution kernel in the channel dimension.

More specifically, to solve the above problem, in the embodiment of the present application, each covariance matrix is first subjected to contrast search space homography, that is, eigenvalues of each position in each covariance matrix in the plurality of covariance matrices are corrected by the following formula to obtain a plurality of corrected covariance matrices; wherein the formula is:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is said plurality of covariance matrices divided by said second

For spatial control of hyper-parameters, e.g. initially set to

The individual covariance matrix and

the distance between the individual covariance matrices.

It can be understood that, by comparing search space syntropy, each covariance matrix can be transferred to an isotropic and differentiated expression space, so as to enhance distribution continuity of feature expression of the three-dimensional input tensor obtained after arrangement along the channel direction and improve feature expression effect of the twisted line segment associated feature map. Thus, the accuracy of quality detection of the cable strand is improved.

In step S160, the corrected covariance matrices are arranged into a three-dimensional input tensor, and then a twisted line segment correlation feature map is obtained through a second convolution neural network with a three-dimensional convolution kernel. That is, the second convolutional neural network using the three-dimensional convolution kernel is used as a feature extractor to capture the correlation features between strand segments of the differences between the local features of the high-dimensional images of different strand segments.

Specifically, in the embodiment of the present application, the layers of the second convolutional neural network with three-dimensional convolutional kernel are used to perform convolutional processing, pooling processing and nonlinear activation processing on input data in forward pass of layers respectively to output the twisted wire segment correlation feature map from the last layer of the second convolutional neural network with three-dimensional convolutional kernel, wherein the input of the first layer of the second convolutional neural network with three-dimensional convolutional kernel is the three-dimensional input tensor.

It can be understood that, for the plurality of corrected covariance matrices, implicit associated feature information on a spatial dimension needs to be more emphasized, and therefore, after the plurality of corrected covariance matrices are arranged into a three-dimensional input tensor, mining of deep associated features is performed by using a second convolutional neural network with a three-dimensional convolutional kernel, so that the second convolutional neural network with the three-dimensional convolutional kernel captures associated features between stranded wire segments of differences between local features of high-dimensional images of different stranded wire segments to obtain the stranded wire segment associated feature map.

In step S170, the stranded wire segment correlation characteristic diagram is passed through a classifier to obtain a classification result, where the classification result is used to indicate whether a cable stranded wire to be detected has a quality defect. Namely, the strand segment association feature map is input into a classification function to obtain a classification function value, wherein the classification function value is the classification result, and the classification result is used for indicating whether the cable strand to be detected has quality defects.

Specifically, in the embodiment of the present application, the classifier is used to process the twisted wire segment association feature map by using the following formula to obtain the classification result, where the formula is:

wherein the content of the first and second substances,

in order to be the result of said classification,

and

is as follows

The weights and bias matrices corresponding to the individual classes,

and associating a characteristic diagram with the stranded wire section.

In summary, the image-based cable strand quality detection method based on the embodiment of the application is clarified, and the method realizes the intellectualization of the cable strand quality detection. Specifically, a plurality of stranded wire section images of a cable stranded wire to be detected are collected along the extending direction of the cable stranded wire to be detected through a camera, then the characteristics of each stranded wire section are extracted through a first convolution neural network with a multi-scale convolution structure through the stranded wire section images, the difference between the characteristics of different stranded wire sections is obtained through covariance calculation, then a second convolution neural network with a three-dimensional convolution kernel is used as a characteristic extractor to capture the correlation characteristics between the differences of the characteristics of the different stranded wire sections so as to obtain a stranded wire section correlation characteristic diagram, and finally a classification result used for indicating whether the cable stranded wire to be detected has quality defects is obtained through a classifier on the stranded wire section correlation characteristic diagram. Therefore, an intelligent detection scheme for the quality of the cable stranded wire is constructed based on the stranded wire image.

Exemplary System

Fig. 5 is a block diagram of an image-based cable strand quality detection system 100 according to an embodiment of the present application. As shown in fig. 5, an image-based cable strand quality detection system 100 according to an embodiment of the present application includes: the image acquisition module 110 is configured to acquire a plurality of stranded wire segment images of a cable stranded wire to be detected along an extending direction of the cable stranded wire to be detected; the first feature extraction module 120 is configured to pass each of the stranded wire segment images of the cable stranded wire to be detected through a first convolutional neural network having a multi-scale convolutional structure to obtain a plurality of multi-scale stranded wire segment feature maps; the pooling processing module 130 is configured to perform global mean pooling along a channel dimension on each of the multiple multi-scale stranded wire segment feature maps respectively to obtain multiple multi-scale stranded wire segment feature vectors; a covariance matrix calculation module 140, configured to calculate a covariance matrix between every two multiscale twisted line segment eigenvectors in the plurality of multiscale twisted line segment eigenvectors to obtain a plurality of covariance matrices; a correcting module 150, configured to correct the eigenvalue of each position in each covariance matrix of the covariance matrices to obtain a plurality of corrected covariance matrices; the second feature extraction module 160 is configured to arrange the corrected covariance matrices into a three-dimensional input tensor and then obtain a twisted-pair segment associated feature map through a second convolutional neural network with a three-dimensional convolutional kernel; and the detection result generation module 170 is configured to pass the stranded wire segment association feature map through a classifier to obtain a classification result, where the classification result is used to indicate whether the cable stranded wire to be detected has a quality defect.

In an example, in the above image-based cable strand quality detection system 100, the first feature extraction module 120 is further configured to: respectively performing the following on input data in the forward transfer process of the layers by using each layer of the first convolutional neural network with the multi-scale convolutional structure: performing convolution processing based on a first convolution kernel on the input data to obtain a first scale convolution characteristic diagram; performing convolution processing based on a second convolution kernel on the input data to obtain a second scale convolution characteristic diagram; performing convolution processing on the input data based on a third convolution kernel to obtain a third scale convolution characteristic diagram; cascading the first scale convolution feature map, the second scale convolution feature map and the third scale convolution feature map to obtain a multi-scale convolution feature map; pooling the multi-scale convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooled feature map to obtain a nonlinear activation feature map; and outputting the final layer of the first convolutional neural network with the multi-scale convolutional structure as the multi-scale twisted line segment characteristic diagram.

In an example, in the above-mentioned image-based cable strand quality detection system 100, the pooling processing module 130 is further configured to: and performing global mean pooling on each characteristic matrix along the channel dimension of each multi-scale stranded wire section characteristic diagram in the multi-scale stranded wire section characteristic diagrams respectively to obtain the multi-scale stranded wire section characteristic vectors.

In an example, in the above-mentioned image-based cable strand quality detection system 100, the correction module 150 is further configured to: correcting the eigenvalue of each position in each covariance matrix in the covariance matrices according to the following formula to obtain a plurality of corrected covariance matrices; wherein the formula is:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is said plurality of covariance matrices divided by said second

the hyper-parameters are controlled for the space.

In an example, in the above-mentioned image-based cable strand quality detection system 100, the second feature extraction module 160 is further configured to: performing convolution processing, pooling processing and nonlinear activation processing on input data in forward pass of layers respectively by using the layers of the second convolutional neural network with the three-dimensional convolutional kernel, so as to output the twisted wire segment correlation feature map by the last layer of the second convolutional neural network with the three-dimensional convolutional kernel, wherein the input of the first layer of the second convolutional neural network with the three-dimensional convolutional kernel is the three-dimensional input tensor.

In an example, in the above-mentioned image-based cable strand quality detection system 100, the detection result generation module 170 is further configured to: processing the twisted wire segment association feature map by using the classifier according to the following formula to obtain the classification result, wherein the formula is as follows:

wherein the content of the first and second substances,

in order to be a result of said classification,

and

is as follows

The weights and bias matrices corresponding to the individual classes,

and associating a characteristic diagram with the stranded wire section.

Here, it may be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described image-based cable strand quality detection system 100 have been described in detail in the above description of the image-based cable strand quality detection method with reference to fig. 1 to 4, and thus, a repeated description thereof will be omitted.

As described above, the image-based cable strand quality detection system 100 according to the embodiment of the present application may be implemented in various terminal devices, such as a server for intelligently detecting the quality of a cable strand. In one example, the image-based cable strand quality detection system 100 according to the embodiment of the present application may be integrated into a terminal device as a software module and/or a hardware module. For example, the image-based cable strand quality detection system 100 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the image-based cable strand quality detection system 100 may also be one of many hardware modules of the terminal device.

Alternatively, in another example, the image-based cable strand quality detection system 100 and the terminal device may be separate devices, and the image-based cable strand quality detection system 100 may be connected to the terminal device through a wired and/or wireless network and transmit the interaction information according to an agreed data format.

Exemplary electronic device

Next, an electronic apparatus according to an embodiment of the present application is described with reference to fig. 6.

As shown in fig. 6, the electronic device 10 includes one or more processors 11 and memory 12.

The processor 11 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.

Memory 12 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium and executed by the processor 11 to implement the image-based cable strand quality detection method of the various embodiments of the present application described above and/or other desired functions. Various contents such as a material of a cable strand may also be stored in the computer-readable storage medium.

In one example, the electronic device 10 may further include: an input device 13 and an output device 14, which are interconnected by a bus system and/or other form of connection mechanism (not shown).

The input device 13 may include, for example, a keyboard, a mouse, and the like.

The output device 14 can output various information including the classification result to the outside. The output devices 14 may include, for example, a display, speakers, a printer, and a communication network and its connected remote output devices, among others.

Of course, for simplicity, only some of the components of the electronic device 10 relevant to the present application are shown in fig. 6, and components such as buses, input/output interfaces, and the like are omitted. In addition, the electronic device 10 may include any other suitable components depending on the particular application.

Exemplary computer program product and computer-readable storage Medium

In addition to the above-described methods and apparatus, embodiments of the present application may also be a computer program product comprising computer program instructions which, when executed by a processor, cause the processor to perform the steps in the functions in the image-based cable strand quality detection method according to various embodiments of the present application described in the above-mentioned "exemplary methods" section of this specification.

The computer program product may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages, for carrying out operations according to embodiments of the present application. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.

Furthermore, embodiments of the present application may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform the steps in the functions in the image-based cable strand quality detection method according to various embodiments of the present application described in the "exemplary method" section above in this specification.

The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The basic principles of the present application have been described above with reference to specific embodiments, but it should be noted that advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.

The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. An image-based cable strand quality detection method is characterized by comprising the following steps: collecting a plurality of stranded wire section images of a cable stranded wire to be detected along the extending direction of the cable stranded wire to be detected; respectively passing each stranded wire section image in a plurality of stranded wire section images of the cable stranded wire to be detected through a first convolution neural network with a multi-scale convolution structure to obtain a plurality of multi-scale stranded wire section characteristic graphs; performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; calculating a covariance matrix between every two multiscale stranded wire section eigenvectors in the multiscale stranded wire section eigenvectors to obtain a plurality of covariance matrices; correcting the eigenvalue of each position in each covariance matrix in the plurality of covariance matrices to obtain a plurality of corrected covariance matrices; arranging the corrected covariance matrixes into a three-dimensional input tensor, and then obtaining a stranded wire section association feature map through a second convolution neural network with a three-dimensional convolution kernel; and the stranded wire section association characteristic graph is processed by a classifier to obtain a classification result, and the classification result is used for indicating whether the cable stranded wire to be detected has quality defects or not.

2. The image-based cable strand quality detection method according to claim 1, wherein the step of passing each strand segment image of a plurality of strand segment images of the cable strand to be detected through a first convolutional neural network having a multi-scale convolutional structure to obtain a plurality of multi-scale strand segment characteristic maps comprises the steps of: respectively performing the following on input data in the forward transfer process of the layers by using each layer of the first convolutional neural network with the multi-scale convolutional structure: performing convolution processing based on a first convolution kernel on the input data to obtain a first scale convolution characteristic diagram; performing convolution processing based on a second convolution kernel on the input data to obtain a second scale convolution characteristic diagram; performing convolution processing on the input data based on a third convolution kernel to obtain a third scale convolution characteristic diagram; cascading the first scale convolution feature map, the second scale convolution feature map and the third scale convolution feature map to obtain a multi-scale convolution feature map; pooling the multi-scale convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooling characteristic map to obtain a nonlinear activation characteristic map; and outputting the final layer of the first convolutional neural network with the multi-scale convolutional structure as the multi-scale twisted line segment characteristic diagram.

3. The image-based cable strand quality detection method of claim 2, wherein the performing global mean pooling along channel dimensions on each of the multiple multi-scale strand segment feature maps to obtain multiple multi-scale strand segment feature vectors comprises: and performing global mean pooling on each characteristic matrix along the channel dimension of each multi-scale stranded wire section characteristic diagram in the multi-scale stranded wire section characteristic diagrams respectively to obtain the multi-scale stranded wire section characteristic vectors.

4. The image-based cable strand quality detection method of claim 3, wherein the correcting the eigenvalues of each position in each of the plurality of covariance matrices to obtain a plurality of corrected covariance matrices comprises: correcting the eigenvalue of each position in each covariance matrix in the covariance matrices according to the following formula to obtain a plurality of corrected covariance matrices; wherein the formula is:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is said plurality of covariance matrices divided by said second

the hyper-parameters are controlled for the space.

5. The image-based cable strand quality detection method according to claim 4, wherein the arranging the corrected covariance matrices into a three-dimensional input tensor is followed by passing through a second convolutional neural network with a three-dimensional convolutional kernel to obtain a strand segment correlation feature map, and the method comprises: performing convolution processing, pooling processing and nonlinear activation processing on input data in forward pass of layers respectively by using the layers of the second convolutional neural network with the three-dimensional convolutional kernel, so as to output the twisted wire segment correlation feature map by the last layer of the second convolutional neural network with the three-dimensional convolutional kernel, wherein the input of the first layer of the second convolutional neural network with the three-dimensional convolutional kernel is the three-dimensional input tensor.

6. The image-based cable strand quality detection method according to claim 5, wherein the step of passing the strand segment association feature map through a classifier to obtain a classification result comprises the steps of: processing the twisted wire segment association feature map by using the classifier according to the following formula to obtain the classification result, wherein the formula is as follows:

wherein the content of the first and second substances,

in order to be the result of said classification,

and

is as follows

The weight and bias matrix corresponding to each class label,

and associating a characteristic diagram with the stranded wire section.

7. An image-based cable strand quality detection system, comprising: the image acquisition module is used for acquiring a plurality of stranded wire section images of the cable stranded wire to be detected along the extension direction of the cable stranded wire to be detected; the first characteristic extraction module is used for enabling each stranded wire section image in the stranded wire section images of the cable stranded wire to be detected to pass through a first convolution neural network with a multi-scale convolution structure respectively to obtain a plurality of multi-scale stranded wire section characteristic graphs; the pooling processing module is used for performing global mean pooling along channel dimensions on each multi-scale stranded wire section characteristic graph in the multi-scale stranded wire section characteristic graphs respectively to obtain a plurality of multi-scale stranded wire section characteristic vectors; the covariance matrix calculation module is used for calculating a covariance matrix between every two multiscale stranded wire section eigenvectors in the multiscale stranded wire section eigenvectors to obtain a plurality of covariance matrices; the correction module is used for correcting the eigenvalue of each position in each covariance matrix in the covariance matrices to obtain a plurality of corrected covariance matrices; the second feature extraction module is used for arranging the corrected covariance matrixes into a three-dimensional input tensor and then obtaining a stranded wire section association feature map through a second convolution neural network with a three-dimensional convolution kernel; and the detection result generation module is used for enabling the stranded wire section correlation characteristic diagram to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the cable stranded wire to be detected has quality defects or not.

8. The image-based cable strand quality detection system of claim 7, wherein the first feature extraction module is further configured to: respectively performing the following on input data in the forward transfer process of the layers by using each layer of the first convolutional neural network with the multi-scale convolutional structure: performing convolution processing on the input data based on a first convolution kernel to obtain a first scale convolution characteristic diagram; performing convolution processing based on a second convolution kernel on the input data to obtain a second scale convolution characteristic diagram; performing convolution processing on the input data based on a third convolution kernel to obtain a third scale convolution characteristic diagram; cascading the first scale convolution feature map, the second scale convolution feature map and the third scale convolution feature map to obtain a multi-scale convolution feature map; pooling the multi-scale convolution characteristic map to obtain a pooled characteristic map; and carrying out nonlinear activation on the pooling characteristic map to obtain a nonlinear activation characteristic map; and outputting the final layer of the first convolutional neural network with the multi-scale convolutional structure as the multi-scale twisted line segment characteristic diagram.

9. The image-based cable strand quality detection system of claim 8, wherein the pooling processing module is further configured to: and performing global mean pooling on each characteristic matrix along the channel dimension of each multi-scale stranded wire section characteristic diagram in the multi-scale stranded wire section characteristic diagrams respectively to obtain the multi-scale stranded wire section characteristic vectors.

10. The image-based cable strand quality detection system of claim 9, wherein the correction module is further configured to: correcting the eigenvalue of each position in each covariance matrix in the covariance matrices according to the following formula to obtain a plurality of corrected covariance matrices; wherein the formula is:

wherein

Is the first of the plurality of covariance matrices

An eigenvalue of each position of the individual covariance matrix, and

is the multiple protocolsDividing said second in a variance matrix

the hyper-parameters are controlled for the space.