Method for detecting surface defects of steel wire rope on production line based on machine vision
By building a frequency domain screening network and enhancing network, the characteristics of the wire rope image are extracted, and the problems of large amount of parameters and low detection accuracy in the existing technology are solved, and fast and accurate defect detection is achieved to meet real-time requirements.
Patent Information
- Application Number
- CN202411342198.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-25
- Publication Date
- 2025-01-14
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the detection of wire rope defects, it is difficult for existing networks to quickly acquire and remember the location of the region of interest for a long time, resulting in large amounts of parameters and insufficient feature extraction, which affects the accuracy of defect detection.
Build a defect detection model including frequency domain screening network, enhanced network, feature extraction network and classification network. Through the combined use of frequency domain screening, enhanced feature extraction and classification network, obtain the filtering frequency domain features, enhanced features and extracting features of wire rope images. , perform defect prediction and model training.
It realizes the active, fast and accurate extraction of the target features of the wire rope image, improves the accuracy of defect detection, reduces the amount of parameters, meets the needs of real-time inspection, replaces manual inspection, and overcomes the problems of low efficiency and long detection cycle.
Smart Images
Figure CN119314097A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of defect detection, and more particularly to a method for detecting surface defects of steel wire ropes on a production line based on machine vision. Background Art
[0002] At present, wire ropes are widely used in various lifting, traction and load-bearing systems, such as hoists, cranes, cableways, cable-stayed bridges, etc. However, due to long-term use and the influence of the external environment, wire ropes are prone to various defects, such as broken wires and wear. These defects will cause the wire ropes to lose their original mechanical properties and safety performance, and may even cause serious accidents. Therefore, wire rope defect detection technology is very important. With the development of computer vision technology, wire rope defect detection methods based on machine vision have gradually become a research hotspot.
[0003] However, the existing network cannot quickly obtain and memorize the location of the region of interest for a long time. Therefore, this type of network architecture needs to use a large number of parameters in the encoder and decoder sub-networks to repeatedly extract the features of the region of interest, which makes it difficult to fully extract the effective features of the wire rope surface, thereby affecting the accuracy of defect detection.
[0004] Therefore, how to efficiently and accurately extract the target features of wire rope images and thus improve the accuracy of wire rope defect detection is a problem that technical personnel in this field need to solve urgently. Summary of the invention
[0005] In view of this, the present invention provides a method for detecting surface defects of wire ropes on a production line based on machine vision, which can accurately extract target features of wire rope images, thereby effectively improving the accuracy of defect detection.
[0006] In order to achieve the above object, the present invention adopts the following technical solution:
[0007] In a first aspect, the present invention provides a method for detecting surface defects of a steel wire rope on a production line based on machine vision, comprising:
[0008] Construct a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network, and a classification network;
[0009] Acquire the wire rope image and input it into the frequency domain screening network to obtain the screening frequency domain features;
[0010] Inputting the filtered frequency domain features and the wire rope image into the enhancement network to obtain enhanced features;
[0011] The enhancement network includes: a candidate region module, a second nonlinear function layer and an upsampling layer;
[0012] The filtered frequency domain features are input into the candidate region module to obtain the image candidate region;
[0013] The image candidate region is sequentially input to the second nonlinear function layer and the upsampling layer for processing to obtain a preprocessed region;
[0014] The pre-processed area is subjected to pixel-by-pixel multiplication operation with the wire rope image to obtain the enhanced feature;
[0015] Inputting the enhanced features into the feature extraction network to obtain extracted features;
[0016] Inputting the extracted features into the classification network to obtain defect prediction results;
[0017] The defect detection model is trained based on the defect prediction result and the defect real label joint loss function to obtain a trained defect detection model;
[0018] The steel wire rope image to be detected is input into the trained defect detection model to obtain the corresponding defect prediction result.
[0019] Preferably, the frequency domain screening network includes: a frequency domain conversion module, a global pooling layer, a first convolution layer, a first activation layer, a second convolution layer, a second activation layer, a third convolution layer and a first nonlinear function layer;
[0020] The steel wire rope image is input into the frequency domain conversion module to obtain the image frequency domain features;
[0021] The image frequency domain features are processed sequentially through the global pooling layer, the first convolution layer, the first activation layer, the second convolution layer, the second activation layer, the third convolution layer and the first nonlinear function layer to obtain screening features;
[0022] The screening feature is fused with the image frequency domain feature to obtain the screening frequency domain feature.
[0023] Preferably, the frequency domain conversion module comprises: a wavelet transform unit, a first normalization layer and a third activation layer;
[0024] The steel wire rope image is input into the wavelet transform unit to obtain wavelet features;
[0025] The wavelet features are sequentially input into the first normalization layers and the third activation layer to obtain the image frequency domain features.
[0026] Preferably, the candidate region module comprises: a fourth convolution layer, a fourth activation layer, a fifth convolution layer, a fifth activation layer and a filtering operator;
[0027] The filtered frequency domain features are sequentially input into the fourth convolution layer, the fourth activation layer, the fifth convolution layer and the fifth activation layer for processing to obtain a significant area;
[0028] The salient region and the filter operator are subjected to a convolution layer operation to obtain the image candidate region.
[0029] Preferably, the feature extraction network comprises: a sixth convolutional layer, a second batch of normalization layers, a sixth activation layer, a seventh convolutional layer, a seventh activation layer, an eighth convolutional layer and an eighth activation layer;
[0030] The enhanced features are sequentially input to the sixth convolutional layer, the second batch normalization layer, the sixth activation layer, the seventh convolutional layer, the seventh activation layer, the eighth convolutional layer and the eighth activation layer for processing to obtain the extracted features.
[0031] Preferably, the classification network comprises: a ninth convolutional layer, a third batch normalization layer, a ninth activation layer, a tenth convolutional layer, a fourth batch normalization layer and a tenth activation layer;
[0032] The extracted features are sequentially input to the ninth convolution layer, the third batch normalization layer, the ninth activation layer, the tenth convolution layer, the fourth batch normalization layer and the tenth activation layer for processing to obtain the defect prediction result.
[0033] Preferably, the loss function is:
[0034] L = -[xlogy + (1-x)log(1-y)];
[0035] Among them, L represents the loss function, x represents the true label value of the defect corresponding to the image, and y represents the predicted output value of the model.
[0036] In a second aspect, the present invention provides a wire rope surface defect detection system on a production line based on machine vision, comprising: a model building module, a screening frequency domain feature acquisition module, an enhanced feature acquisition module, a prediction module, a model training module and a result output module;
[0037] The model building module is used to build a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network and a classification network;
[0038] The filtering frequency domain feature acquisition module is used to acquire the wire rope image and input it into the frequency domain filtering network to obtain the filtering frequency domain feature;
[0039] The enhanced feature acquisition module is used to input the filtered frequency domain features and the wire rope image into the enhanced network to obtain enhanced features;
[0040] The prediction module is used to input the enhanced features into the feature extraction network to obtain extracted features; and input the extracted features into the classification network to obtain defect prediction results;
[0041] The model training module is used to train the defect detection model based on the defect prediction result and the defect real label joint loss function to obtain a trained defect detection model;
[0042] The result output module is used to input the steel wire rope image to be detected into the trained defect detection model to obtain the corresponding defect prediction result.
[0043] In a third aspect, the present invention provides a computer device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;
[0044] Memory, used to store computer programs;
[0045] The processor, when used to execute the program stored in the memory, can implement a method for detecting surface defects of wire ropes on a production line based on machine vision as described in the first aspect.
[0046] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for detecting surface defects of wire ropes on a production line based on machine vision, which can accurately extract target features of wire rope images, thereby effectively improving the accuracy of defect detection and having the following beneficial effects:
[0047] 1. The method of the present invention can actively, quickly and accurately extract high-dimensional features of targets in wire rope images. The method has strong interpretability and low parameter requirements, and the hardware conditions of the operating platform are much lower than those of general deep learning target detection methods.
[0048] 2. The method proposed in the present invention can replace manual inspection, determine whether the appearance of the wire rope is abnormal, and detect in real time whether there is damage on the surface of the wire rope. It overcomes the problems of low efficiency, long detection cycle and inability to detect in time of manual inspection, and can meet the real-time requirements.
[0049] 3. The present invention can accurately extract target features through the enhanced network and feature extraction network, and input the target features into the classification network, making full use of the feature information, and can accurately identify and locate potential defects in the wire rope image, thereby effectively improving the accuracy of defect detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0051] Figure 1 A flow chart of a method for detecting surface defects of wire ropes on a production line based on machine vision provided by the present invention.
[0052] Figure 2 This is a schematic diagram of the defect detection model structure provided by the present invention.
[0053] Figure 3 This is a schematic diagram of the frequency domain screening network structure provided by the present invention.
[0054] Figure 4 This is a schematic diagram of the enhanced network structure provided by the present invention.
[0055] Figure 5 This is a schematic structural diagram of a wire rope surface defect detection system on a production line based on machine vision provided by the present invention. DETAILED DESCRIPTION
[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0057] Example 1
[0058] like Figure 1 As shown, the embodiment of the present invention discloses a method for detecting surface defects of a steel wire rope on a production line based on machine vision, comprising:
[0059] Construct a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network, and a classification network;
[0060] Obtain the wire rope image and input it into the frequency domain screening network to obtain the screening frequency domain features;
[0061] Input the filtered frequency domain features and the wire rope image into the enhancement network to obtain enhanced features;
[0062] Input the enhanced features into the feature extraction network to obtain the extracted features;
[0063] Input the extracted features into the classification network to obtain defect prediction results;
[0064] The defect detection model is trained based on the defect prediction results and the defect real label joint loss function to obtain a trained defect detection model;
[0065] The wire rope image to be inspected is input into the trained defect detection model to obtain the corresponding defect prediction result.
[0066] Example 2
[0067] The embodiment of the present invention discloses a method for detecting surface defects of a steel wire rope on a production line based on machine vision, comprising:
[0068] like Figure 2 As shown, a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network and a classification network is constructed.
[0069] Get the wire rope image and input it into the frequency domain filtering network to obtain the filtered frequency domain features:
[0070] Preferably, the wire rope image acquisition method is:
[0071] Set a suitable camera position on the wire rope production line, and use a high-resolution industrial camera to collect images in a multi-angle and multi-directional manner to obtain more comprehensive wire rope surface information.
[0072] Preferably, before inputting into the frequency domain screening network, image preprocessing is also included: removing noise in the image, enhancing useful information in the image, such as contrast, brightness, etc., to make the defect features more obvious.
[0073] Preferably, Figure 3 As shown, the frequency domain screening network includes: a frequency domain conversion module, a global pooling layer, a first convolution layer, a first activation layer, a second convolution layer, a second activation layer, a third convolution layer and a first nonlinear function layer;
[0074] The wire rope image is input into the frequency domain conversion module to obtain the image frequency domain features;
[0075] The image frequency domain features are processed in sequence through the global pooling layer, the first convolution layer, the first activation layer, the second convolution layer, the second activation layer, the third convolution layer and the first nonlinear function layer to obtain the screening features;
[0076] The screening features are fused with the image frequency domain features to obtain the screening frequency domain features.
[0077] Preferably, in this embodiment, the first convolution layer and the second convolution layer both use 1×1 two-dimensional convolution, the first nonlinear function layer uses the nonlinear transformation Sigmoid function, and the first activation layer and the second activation layer use the nonlinear transformation SiLU function.
[0078] Preferably, the frequency domain conversion module comprises: a wavelet transform unit, a first normalization layer and a third activation layer;
[0079] The wire rope image is input into the wavelet transform unit to obtain the wavelet features;
[0080] The wavelet features are sequentially input into the first normalization layers and the third activation layer to obtain the image frequency domain features.
[0081] Preferably, in this embodiment, the wavelet transform unit performs: Harr wavelet transform or DB2 wavelet transform, and uses wavelet transform to perform row and column decomposition, which can quickly extract the image frequency domain features required for locating the region of interest of the image.
[0082] The filtered frequency domain features and wire rope images are input into the enhancement network to obtain the enhanced features:
[0083] Preferably, Figure 4 As shown, the enhanced network includes: a candidate region module, a second nonlinear function layer and an upsampling layer;
[0084] Filter the frequency domain features and input them into the candidate region module to obtain the candidate region of the image;
[0085] The image candidate regions are sequentially input into the second nonlinear function layer and the upsampling layer for processing to obtain the preprocessed regions;
[0086] The preprocessed area is multiplied pixel by pixel with the wire rope image to obtain an enhanced image.
[0087] Preferably, the candidate region module includes: a fourth convolution layer, a fourth activation layer, a fifth convolution layer, a fifth activation layer and a filtering operator;
[0088] The filtered frequency domain features are sequentially input into the fourth convolution layer, the fourth activation layer, the fifth convolution layer, and the fifth activation layer for processing to obtain the significant areas;
[0089] The salient region is subjected to convolutional layer operation with the filter operator to obtain the candidate region of the image.
[0090] Preferably, in this embodiment, the fourth convolution layer adopts 3×3 two-dimensional convolution, the fifth convolution layer adopts 1×1 two-dimensional convolution, and the fourth activation layer and the fifth activation layer both adopt nonlinear transformation SiLU function.
[0091] Input the enhanced features into the feature extraction network to obtain the extracted features:
[0092] Preferably, the feature extraction network includes: a sixth convolutional layer, a second batch of normalization layers, a sixth activation layer, a seventh convolutional layer, a seventh activation layer, an eighth convolutional layer and an eighth activation layer;
[0093] The enhanced features are sequentially input to the sixth convolutional layer, the second batch normalization layer, the sixth activation layer, the seventh convolutional layer, the seventh activation layer, the eighth convolutional layer and the eighth activation layer for processing to obtain the extracted features.
[0094] Preferably, the sixth convolution layer, the seventh convolution layer and the eighth convolution layer all use 3×3 two-dimensional convolution, the second batch normalization layer uses a batch normalization function, and the sixth activation layer, the seventh activation layer and the eighth activation layer all use a nonlinear transformation SiLU function.
[0095] Input the extracted features into the classification network to obtain the defect prediction results:
[0096] Preferably, the classification network comprises: a ninth convolutional layer, a third batch normalization layer, a ninth activation layer, a tenth convolutional layer, a fourth batch normalization layer and a tenth activation layer;
[0097] The extracted features are sequentially input into the ninth convolution layer, the third batch normalization layer, the ninth activation layer, the tenth convolution layer, the fourth batch normalization layer and the tenth activation layer for processing to obtain the defect prediction results.
[0098] Preferably, in this embodiment, the convolution kernel size of the ninth convolution layer is 3, the step size is 1, and the filling factor is 1; the convolution kernel size of the tenth convolution layer is 3, the step size is 1, and the filling factor is 1.
[0099] The defect detection model is trained based on the defect prediction results and the defect real label joint loss function to obtain the trained defect detection model:
[0100] Preferably, the loss function is:
[0101] L = -[xlogy + (1-x)log(1-y)]
[0102] Among them, L represents the loss function, x represents the true label value of the defect corresponding to the image, and y represents the predicted output value of the model.
[0103] Preferably, the defect detection model is trained until the loss no longer decreases, which means that the defect detection model training has reached stability, the training process is completed, and a trained defect detection model is obtained.
[0104] The wire rope image to be inspected is input into the trained defect detection model to obtain the corresponding defect prediction result.
[0105] Example 3
[0106] like Figure 5As shown, a wire rope surface defect detection system on a production line based on machine vision includes: a model building module, a screening frequency domain feature acquisition module, an enhanced feature acquisition module, a prediction module, a model training module and a result output module;
[0107] A model building module for building a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network, and a classification network;
[0108] A filtering frequency domain feature acquisition module is used to acquire the wire rope image and input it into the frequency domain filtering network to obtain the filtering frequency domain feature;
[0109] An enhanced feature acquisition module is used to input the filtered frequency domain features and the wire rope image into the enhanced network to obtain enhanced features;
[0110] The prediction module is used to input the enhanced features into the feature extraction network to obtain the extracted features; and input the extracted features into the classification network to obtain the defect prediction results;
[0111] The model training module is used to train the defect detection model based on the defect prediction results and the defect real label joint loss function to obtain a trained defect detection model;
[0112] The result output module is used to input the wire rope image to be detected into the trained defect detection model to obtain the corresponding defect prediction result.
[0113] Preferably, the process of implementing a wire rope surface defect detection system on a production line based on machine vision disclosed in the present invention corresponds one-to-one to the process of implementing the above-mentioned wire rope surface defect detection method on a production line based on machine vision.
[0114] Example 4
[0115] A computer device comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;
[0116] Memory, used to store computer programs;
[0117] The processor, when used to execute the program stored in the memory, can implement a method for detecting surface defects of wire ropes on a production line based on machine vision as described in any one of Embodiments 1 or 2.
[0118] The electronic device may include: a processor, a communications interface, a memory, and a communications bus, wherein the processor, the communications interface, and the memory communicate with each other via the communications bus. The processor may call the logic instructions in the memory to execute a method for detecting surface defects of wire ropes on a production line based on machine vision in Embodiment 1 or 2:
[0119] Construct a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network, and a classification network;
[0120] Obtain the wire rope image and input it into the frequency domain screening network to obtain the screening frequency domain features;
[0121] Input the filtered frequency domain features and the wire rope image into the enhancement network to obtain enhanced features;
[0122] Input the enhanced features into the feature extraction network to obtain the extracted features;
[0123] Input the extracted features into the classification network to obtain defect prediction results;
[0124] The defect detection model is trained based on the defect prediction results and the defect real label joint loss function to obtain a trained defect detection model;
[0125] The wire rope image to be inspected is input into the trained defect detection model to obtain the corresponding defect prediction result.
[0126] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.
[0127] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method for detecting surface defects of wire ropes on a production line based on machine vision, which can accurately extract target features of wire rope images, thereby effectively improving the accuracy of defect detection and having the following beneficial effects:
[0128] 1. The method of the present invention can actively, quickly and accurately extract high-dimensional features of targets in wire rope images. The method has strong interpretability and low parameter requirements, and the hardware conditions of the operating platform are much lower than those of general deep learning target detection methods.
[0129] 2. The method proposed in the present invention can replace manual inspection, determine whether the appearance of the wire rope is abnormal, and detect in real time whether there is damage on the surface of the wire rope. It overcomes the problems of low efficiency, long detection cycle and inability to detect in time of manual inspection, and can meet the real-time requirements.
[0130] 3. The present invention can accurately extract target features through the enhanced network and feature extraction network, and input the target features into the classification network, making full use of the feature information to accurately describe whether defects occur, thereby effectively improving the accuracy of defect detection.
[0131] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0132] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for detecting surface defects of steel wire ropes on a production line based on machine vision, characterized in that: include: Construct a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network, and a classification network; Acquire the wire rope image and input it into the frequency domain screening network to obtain the screening frequency domain features; Inputting the filtered frequency domain features and the wire rope image into the enhancement network to obtain enhanced features; The enhancement network includes: a candidate region module, a second nonlinear function layer and an upsampling layer; The filtered frequency domain features are input into the candidate region module to obtain the image candidate region; The image candidate region is sequentially input to the second nonlinear function layer and the upsampling layer for processing to obtain a preprocessed region; The pre-processed area is subjected to pixel-by-pixel multiplication operation with the wire rope image to obtain the enhanced feature; Inputting the enhanced features into the feature extraction network to obtain extracted features; Inputting the extracted features into the classification network to obtain defect prediction results; The defect detection model is trained based on the defect prediction result and the defect real label joint loss function to obtain a trained defect detection model; The steel wire rope image to be detected is input into the trained defect detection model to obtain the corresponding defect prediction result.
2. According to the method for detecting surface defects of steel wire ropes on a production line based on machine vision in claim 1, it is characterized in that: The frequency domain screening network includes: a frequency domain conversion module, a global pooling layer, a first convolution layer, a first activation layer, a second convolution layer, a second activation layer, a third convolution layer and a first nonlinear function layer; The steel wire rope image is input into the frequency domain conversion module to obtain the image frequency domain features; The image frequency domain features are processed sequentially through the global pooling layer, the first convolution layer, the first activation layer, the second convolution layer, the second activation layer, the third convolution layer and the first nonlinear function layer to obtain screening features; The screening feature is fused with the image frequency domain feature to obtain the screening frequency domain feature.
3. According to the method for detecting surface defects of steel wire ropes on a production line based on machine vision in claim 2, it is characterized in that: The frequency domain conversion module includes: a wavelet transform unit, a first normalization layer and a third activation layer; The steel wire rope image is input into the wavelet transform unit to obtain wavelet features; The wavelet features are sequentially input into the first normalization layers and the third activation layer to obtain the image frequency domain features.
4. According to the method for detecting surface defects of steel wire ropes on a production line based on machine vision in claim 1, it is characterized in that: The candidate region module comprises: a fourth convolution layer, a fourth activation layer, a fifth convolution layer, a fifth activation layer and a filtering operator; The filtered frequency domain features are sequentially input into the fourth convolution layer, the fourth activation layer, the fifth convolution layer and the fifth activation layer for processing to obtain a significant area; The salient region and the filter operator are subjected to a convolution layer operation to obtain the image candidate region.
5. According to the method for detecting surface defects of steel wire ropes on a production line based on machine vision in claim 1, it is characterized in that: The feature extraction network includes: a sixth convolutional layer, a second batch of normalization layers, a sixth activation layer, a seventh convolutional layer, a seventh activation layer, an eighth convolutional layer and an eighth activation layer; The enhanced features are sequentially input to the sixth convolutional layer, the second batch normalization layer, the sixth activation layer, the seventh convolutional layer, the seventh activation layer, the eighth convolutional layer and the eighth activation layer for processing to obtain the extracted features.
6. The method for detecting surface defects of steel wire ropes on a production line based on machine vision according to claim 1 is characterized in that: The classification network includes: a ninth convolutional layer, a third batch normalization layer, a ninth activation layer, a tenth convolutional layer, a fourth batch normalization layer and a tenth activation layer; The extracted features are sequentially input to the ninth convolution layer, the third batch normalization layer, the ninth activation layer, the tenth convolution layer, the fourth batch normalization layer and the tenth activation layer for processing to obtain the defect prediction result.
7. The method for detecting surface defects of steel wire ropes on a production line based on machine vision according to claim 1 is characterized in that: The loss function is: L = -[xlogy + (1-x)log(1-y)]; Among them, L represents the loss function, x represents the true label value of the defect corresponding to the image, and y represents the predicted output value of the model.
8. A machine vision-based wire rope surface defect detection system on a production line, using the method described in any one of claims 1 to 7, characterized in that: include: Model building module, frequency domain feature acquisition module, enhanced feature acquisition module, prediction module, model training module and result output module; The model building module is used to build a defect detection model including a frequency domain screening network, an enhancement network, a feature extraction network and a classification network; The filtering frequency domain feature acquisition module is used to acquire the wire rope image and input it into the frequency domain filtering network to obtain the filtering frequency domain feature; The enhanced feature acquisition module is used to input the filtered frequency domain features and the wire rope image into the enhanced network to obtain enhanced features; The prediction module is used to input the enhanced features into the feature extraction network to obtain extracted features; Inputting the extracted features into the classification network to obtain defect prediction results; The model training module is used to train the defect detection model based on the defect prediction result and the defect real label joint loss function to obtain a trained defect detection model; The result output module is used to input the steel wire rope image to be detected into the trained defect detection model to obtain the corresponding defect prediction result.
9. A computer device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory, used to store computer programs; The processor, when used to execute the program stored in the memory, can implement a method for detecting surface defects of wire ropes on a production line based on machine vision as described in any one of claims 1 to 7.