Wire core tool mark texture elimination method based on deep learning and detection system

Abstract

The invention discloses a wire core tool mark texture elimination method based on deep learning and a detection system. The method comprises the following steps: constructing a cable core identification and detection assembly based on machine vision, collecting a cable core data set through the cable core identification and detection assembly, extracting target cable core data containing tool mark textures from the cable core data set, performing tool mark texture feature extraction, and constructing a denoising coding and decoding network model based on deep learning according to the tool mark texture features; training a denoising coding and decoding network model based on deep learning through the wire core data set, and verifying the elimination effect of the model on the tool mark texture; and determining model validity and model precision according to a verification result, performing model qualification judgment, and if the judgment is qualified, performing elimination processing on a subsequent line core image with a tool mark by using the de-noising coding and decoding network model. According to the method, conductor contour information of the wire core image can be reserved, image dimension reduction is not carried out, edge information of the wire core image is reserved to the maximum extent, meanwhile, tool mark textures are effectively segmented, and mapping from the image with the tool marks to the image without the tool marks is achieved.

Description

Technical Field

[0001] This invention relates to the fields of machine vision and image processing technology, and in particular to a method and detection system for eliminating wire core knife marks texture based on deep learning. Background Technology

[0002] Currently, during the manufacturing process of cables, metal wires, and other conductors, cutting tools leave periodic or random tool marks (such as spiral marks and scratches) on the surface. These textures interfere with subsequent surface defect detection (such as cracks and pits). Traditional conductor detection relies on manual methods or simple image processing, which are subject to tool mark interference, leading to inaccurate conductor contour extraction and a high false alarm rate. Existing denoising methods (such as traditional filtering algorithms) struggle to distinguish between tool mark noise and the actual conductor structure, easily resulting in the loss of edge information. This leads to low accuracy in conductor detection, reducing work efficiency and practicality. Summary of the Invention

[0003] To address the aforementioned problems, this invention provides a deep learning-based method and detection system for eliminating knife-mark textures in wire cores. This solves the problems mentioned in the background section, where traditional wire core detection relies on manual labor or simple image processing, resulting in interference from knife-mark textures, inaccurate conductor contour extraction, and a high false alarm rate. Existing denoising methods (such as traditional filtering algorithms) struggle to distinguish between knife-mark noise and the actual conductor structure, easily leading to the loss of edge information. This results in low accuracy for wire core detection, reducing work efficiency and practicality.

[0004] A deep learning-based method for eliminating tool marks on wire cores includes the following steps:

[0005] A machine vision-based core identification and detection component is constructed and a core dataset is collected using it. Target core data containing knife mark textures is then extracted from the core dataset.

[0006] Based on the target wire core data, knife mark texture features are extracted, and a deep learning-based denoising encoding and decoding network model is constructed based on the knife mark texture features;

[0007] A deep learning-based denoising codec network model was trained using a wire core dataset, and the model's effectiveness in eliminating knife mark textures was verified.

[0008] The validity and accuracy of the model are determined based on the verification results, and the model is deemed qualified. If the model is deemed qualified, the noise reduction encoding and decoding network model is used to remove the knife-marked wire core images in the subsequent process.

[0009] Preferably, the step of constructing a machine vision-based wire core recognition and detection component and collecting a wire core dataset using it, and extracting target wire core data containing knife mark textures from the wire core dataset, includes:

[0010] Obtain the reflection parameters of the wire core, determine the lens refraction angle based on the reflection parameters, and configure the relative position parameters of the coaxial light source and the diffuse reflector according to the lens refraction angle;

[0011] Determine the edge structure parameters of the cable core, determine the pixel error based on the edge structure parameters, and determine the resolution requirement based on the pixel error;

[0012] Pixel configuration parameters are selected based on resolution requirements. A target specification monochrome CCD camera is selected based on the pixel configuration parameters. A machine vision-based core recognition and detection component is generated by using the target specification monochrome CCD camera, a coaxial light source, a diffuse reflector, and a lens.

[0013] Multiple qualified cable core images are acquired by the core identification and detection component and a core dataset is generated to determine the identification parameters of knife mark texture.

[0014] Based on the recognition parameters, target cable core images containing knife mark textures are extracted from the core dataset and target core data is generated.

[0015] Preferably, the step of extracting knife mark texture features based on target wire core data and constructing a deep learning-based denoising encoding / decoding network model based on the knife mark texture features includes:

[0016] Obtain a cross-sectional diagram of the wire core with knife-mark texture based on the target wire core data, and retrieve the conductor structure diagram of the cable core from the database;

[0017] The image features of the cross-sectional image of the wire core and the structural image of the cable core are extracted and compared to determine the differences in image features. Based on the differences in image features, the texture description features of the knife marks are determined.

[0018] The mapping nonlinear parameters between the knife mark texture image and the knife mark removed texture image are determined based on the knife mark texture description features;

[0019] The architecture of a deep learning-based autoencoder denoising model is selected by mapping nonlinear parameters between the knife-mark texture image and the image with the knife-mark removed, and a denoising encoder-decoder network model is constructed.

[0020] Preferably, the step of training a deep learning-based denoising codec network model using a wire core dataset and verifying the model's effectiveness in eliminating knife mark textures includes:

[0021] Multiple original wire core images in the wire core dataset are rotated to obtain multiple incremental wire core images. The incremental wire core images and the original wire core images are integrated and proportionally allocated into the model training set, the model test set, and the model validation set.

[0022] In the model training set, a preset batch of wire core images are randomly selected to train the deep learning-based denoising encoding and decoding network model. The model output image is collected and compared with the image with knife mark texture to determine the model loss.

[0023] The network model parameters are optimized based on the model loss until the model converges. The converged denoising codec network model is tested using a model test set and validated using a model validation set.

[0024] The success rate of the denoising codec network model in removing knife marks was determined based on the validation data, and the effectiveness of the denoising codec network model in removing knife marks was determined based on the success rate.

[0025] Preferably, the step of determining the model's effectiveness and accuracy based on the verification results and performing a model qualification judgment, and if the judgment is qualified, using a denoising encoding / decoding network model to remove the knife-marked wire core image in subsequent processing, includes:

[0026] Based on the verification results, the correct and incorrect segmentation sample data of the denoising codec network model after the removal of knife mark texture were determined.

[0027] The effectiveness and accuracy of the denoising encoder-decoder network model are determined based on correctly segmented and incorrectly segmented sample data.

[0028] Based on preset effectiveness and accuracy indicators, the denoising codec network model is judged to be qualified according to the model effectiveness and model accuracy.

[0029] If the result is deemed satisfactory, subsequent images of wire cores with knife marks are acquired and preprocessed. A denoising codec network model is then used to remove the knife mark texture from the preprocessed images of the wire cores with knife marks, and an output image is obtained.

[0030] Preferably, determining the correctly segmented and incorrectly segmented sample data of the denoising codec network model for the knife mark texture removal based on the verification results includes:

[0031] Based on the image features of samples before and after knife mark texture removal, a watershed segmentation threshold is set, and multiple model output images are obtained based on the verification results.

[0032] For each model output image, extract the core contour and calculate the current number of cores and the area of ​​a single conductor within the extracted contour area;

[0033] The segmented conductor contour edge parameters corresponding to each model output image are determined based on the current number of conductors within the contour region and the area of ​​a single conductor.

[0034] The current watershed segmentation value of each model output image is determined based on the edge parameters of the segmented conductor contour. The current watershed segmentation value is compared with the watershed segmentation threshold. The first model output image with the same value is taken as the correctly segmented sample data, and the second model output image with different value is taken as the incorrectly segmented sample data.

[0035] A deep learning-based system for detecting and eliminating tool marks on wire cores, the system comprising:

[0036] The extraction module is used to build a machine vision-based wire core recognition and detection component and collect a wire core dataset to extract target wire core data containing knife mark texture from the wire core dataset.

[0037] The module is used to extract knife mark texture features based on target wire core data and construct a deep learning-based denoising encoding and decoding network model based on the knife mark texture features.

[0038] The verification module is used to train a deep learning-based denoising codec network model using a wire core dataset and to verify the model's effectiveness in eliminating knife mark textures.

[0039] The elimination module is used to determine the effectiveness and accuracy of the model based on the verification results and to make a model qualification judgment. If the model is qualified, the noise reduction codec network model is used to eliminate the knife-marked wire core images in the future.

[0040] Preferably, the extraction module includes:

[0041] The configuration submodule is used to obtain the reflection parameters of the wire core, determine the lens refraction angle based on the reflection parameters, and configure the relative position parameters of the coaxial light source and the diffuse reflector according to the lens refraction angle.

[0042] The first determination submodule is used to determine the edge structure parameters of the cable core, determine the pixel error based on the edge structure parameters, and determine the resolution requirement based on the pixel error.

[0043] The generation submodule is used to select pixel configuration parameters based on resolution requirements, select a target specification monochrome CCD camera based on the pixel configuration parameters, and generate a machine vision-based core recognition and detection component through the target specification monochrome CCD camera, coaxial light source, diffuse reflector, and lens.

[0044] The second determination submodule is used to acquire multiple qualified cable core images through the core identification and detection component and generate a core dataset to determine the identification parameters of the knife mark texture.

[0045] The extraction submodule is used to extract target cable core images containing knife mark textures from the core dataset based on recognition parameters and generate target core data.

[0046] Preferably, the building module includes:

[0047] The acquisition submodule is used to obtain a cross-sectional diagram of the wire core with knife-mark texture based on the target wire core data, and to obtain the conductor structure diagram of the cable core from the database;

[0048] The third determination submodule is used to extract the image features of the wire core cross-section diagram and the cable wire core conductor structure diagram respectively, compare them to determine the difference image features, and determine the knife mark texture description features based on the difference image features.

[0049] The fourth determination submodule is used to determine the mapping nonlinear parameters of the knife mark texture image and the removed knife mark texture image based on the knife mark texture description features;

[0050] A submodule is constructed to select a deep learning-based autoencoder denoising model architecture and construct a denoising encoder-decoder network model by mapping nonlinear parameters between the knife-mark texture image and the image with the knife-mark texture removed.

[0051] Preferably, the verification module includes:

[0052] The allocation submodule is used to rotate multiple original core images in the core dataset to obtain multiple incremental core images, integrate the incremental core images and original core images, and allocate them proportionally as a model training set, a model test set, and a model validation set.

[0053] The fifth determination submodule is used to randomly select a preset batch of wire core images in the model training set to train the deep learning-based denoising encoding and decoding network model, collect the model output image and compare it with the knife mark texture image to determine the model loss.

[0054] The validation submodule is used to optimize the network model parameters based on the model loss until the model converges, test the converged denoising codec network model with the model test set, and validate the converged denoising codec network model with the model validation set.

[0055] The sixth determination submodule is used to determine the success rate of the denoising codec network model in removing knife marks based on the verification data, and to determine the effect of the denoising codec network model on removing knife marks based on the success rate.

[0056] The elimination module includes:

[0057] The seventh determination submodule is used to determine the correct and incorrect segmentation sample data of the denoising codec network model after the removal of knife marks and textures, based on the verification results.

[0058] The eighth determination submodule is used to determine the model validity and model accuracy of the denoising encoder-decoder network model based on correctly segmented sample data and incorrectly segmented sample data;

[0059] The judgment submodule is used to determine the pass / fail status of the denoising codec network model based on the model's validity and accuracy according to preset validity and accuracy indicators.

[0060] The elimination submodule is used to acquire subsequent images of wire cores with knife marks if the judgment is qualified, and preprocess them. Then, the knife mark texture is eliminated by the preprocessed subsequent images of wire cores with knife marks through a denoising codec network model to obtain the output image.

[0061] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0062] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0063] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0064] Figure 1 A flowchart illustrating the workflow of a deep learning-based method for eliminating wire core tool marks provided by this invention;

[0065] Figure 2 Another flowchart of a deep learning-based method for eliminating wire core tool marks texture provided by the present invention;

[0066] Figure 3 A schematic diagram of the structure of a deep learning-based wire core tool mark texture removal and detection system provided by the present invention;

[0067] Figure 4 This is a schematic diagram of the structure of the building module in a deep learning-based wire core tool mark texture elimination system provided by the present invention. Detailed Implementation

[0068] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0069] Currently, during the manufacturing process of cables, metal wires, and other conductors, cutting tools leave periodic or random tool marks (such as spiral marks, scratches, etc.) on the surface. These textures interfere with subsequent surface defect detection (such as cracks, pits). Traditional conductor detection relies on manual methods or simple image processing, which are subject to tool mark texture interference, leading to inaccurate conductor contour extraction and a high false alarm rate. Existing denoising methods (such as traditional filtering algorithms) struggle to distinguish between tool mark noise and the actual conductor structure, easily resulting in the loss of edge information, thus causing low accuracy in conductor detection and reducing work efficiency and practicality. To address these issues, this embodiment discloses a deep learning-based method for eliminating conductor tool mark textures.

[0070] A deep learning-based method for eliminating tool marks on wire cores, such as... Figure 1 As shown, it includes the following steps:

[0071] Step S101: Construct a machine vision-based wire core recognition and detection component and collect a wire core dataset using it, then extract target wire core data containing knife mark textures from the wire core dataset.

[0072] Step S102: Extract knife mark texture features based on target wire core data and construct a deep learning-based denoising encoding / decoding network model based on knife mark texture features;

[0073] Step S103: Train a deep learning-based denoising codec network model using the wire core dataset and verify the model's effect on eliminating knife mark textures;

[0074] Step S104: Determine the model's validity and accuracy based on the verification results and make a model qualification judgment. If the judgment is qualified, use the denoising codec network model to eliminate the knife-marked wire core image in the subsequent process.

[0075] The working principle of the above technical solution is as follows: a machine vision-based wire core recognition and detection component is constructed and a wire core dataset is collected through it. Target wire core data containing knife mark textures is extracted from the wire core dataset. Knife mark texture features are extracted based on the target wire core data, and a deep learning-based denoising codec network model is constructed based on the knife mark texture features. The deep learning-based denoising codec network model is trained using the wire core dataset, and the model's effect on eliminating knife mark textures is verified. The effectiveness and accuracy of the model are determined based on the verification results, and a model qualification judgment is made. If the judgment is qualified, the denoising codec network model is used to remove knife mark textures from subsequent wire core images.

[0076] The beneficial effects of the above technical solution are as follows: By constructing a machine vision-based wire core recognition and detection component, the acquired wire core images can be ensured to be high-fidelity images, guaranteeing image quality. Furthermore, by constructing a denoising encoding and decoding network model, a fully convolutional approach can be adopted in the convolution and deconvolution processes, reducing pooling layers and facilitating the preservation of conductor contour information in the wire core image. Without image dimensionality reduction, the edge information of the wire core image is preserved to the maximum extent. Simultaneously, knife mark textures are effectively segmented, achieving a mapping from images with knife marks to images without knife marks. This improves the detection accuracy, work efficiency, and practicality of wire cores. It solves the problem mentioned in existing technologies where traditional wire core detection relies on manual or simple image processing, resulting in interference from knife mark textures, leading to inaccurate conductor contour extraction and a high false alarm rate. Existing denoising methods (such as traditional filtering algorithms) struggle to distinguish between knife mark noise and the real conductor structure, easily causing the loss of edge information, resulting in low accuracy in wire core detection and reduced work efficiency and practicality.

[0077] In this embodiment, after extracting the target wire core data containing the knife mark texture from the wire core dataset, the method further includes:

[0078] Image enhancement preprocessing is performed on each image of the target wire core data containing knife marks, and the shape features of the knife marks in each image of the preprocessed wire core containing knife marks are detected and identified;

[0079] Based on the shape features of the knife marks, all preprocessed images of wire cores containing knife marks are classified, and multiple shape description vectors of the knife mark texture are determined based on the classification results.

[0080] Obtain the texture statistics parameters of each shape description vector, and determine the global statistics vector and local edge statistics vector of each shape description vector based on the texture statistics parameters;

[0081] The strong pixel recognition features and weak pixel recognition features of each shape description vector are determined based on the global statistical vector and the local edge statistical vector.

[0082] Based on strong pixel recognition features and weak pixel recognition features, the classification of each shape description vector includes filtering of knife-marked wire core images;

[0083] Based on the filtering results, obtain the matching knife mark core image for each shape description vector, and determine the regional self-similarity between the pairwise matching knife mark core images for each shape description vector;

[0084] The texture connectivity domain of each shape description vector is determined based on the region self-similarity, and the texture morphology rules of each shape description vector are determined based on the two-dimensional spatial description of the texture connectivity domain.

[0085] Based on the texture morphology rules, the aggregated morphological feature vector of each shape description vector is determined, and the knife mark texture feature extraction strategy of each shape description vector is generated based on the aggregated morphological feature vector.

[0086] The knife mark texture feature extraction strategy is based on the target wire core data and is performed using a knife mark texture feature extraction strategy for each shape description vector.

[0087] The beneficial effects of the above technical solution are as follows: by screening image samples of the target wire core data, the influence of low reference value or high similarity knife mark texture images can be removed, thereby ensuring the high quality and reliability of the sample data. Furthermore, by determining the aggregated morphological feature vector of each shape description vector, the corresponding morphological features can be accurately located and generated based on the texture region of each shape knife mark image, thereby formulating a knife mark texture feature extraction strategy to ensure the reliability, stability and adaptability of the extraction results, and further improving practicality.

[0088] In one embodiment, the construction of a machine vision-based wire core recognition and detection component, and the acquisition of a wire core dataset thereon, along with the extraction of target wire core data containing knife mark textures from the wire core dataset, includes:

[0089] Obtain the reflection parameters of the wire core, determine the lens refraction angle based on the reflection parameters, and configure the relative position parameters of the coaxial light source and the diffuse reflector according to the lens refraction angle;

[0090] Determine the edge structure parameters of the cable core, determine the pixel error based on the edge structure parameters, and determine the resolution requirement based on the pixel error;

[0091] Pixel configuration parameters are selected based on resolution requirements. A target specification monochrome CCD camera is selected based on the pixel configuration parameters. A machine vision-based core recognition and detection component is generated by using the target specification monochrome CCD camera, a coaxial light source, a diffuse reflector, and a lens.

[0092] Multiple qualified cable core images are acquired by the core identification and detection component and a core dataset is generated to determine the identification parameters of knife mark texture.

[0093] Based on the recognition parameters, target cable core images containing knife mark textures are extracted from the core dataset and target core data is generated.

[0094] The beneficial effects of the above technical solution are as follows: by using a target-specification black and white CCD camera with a coaxial light source, a diffuse reflector, and a lens to generate a machine vision-based core identification and detection component, the interference of reflections can be overcome and the high quality of the acquired core images can be ensured. Furthermore, by screening images based on the identification parameters of the knife mark texture, images containing knife marks can be screened accurately and without omission, thus improving the screening accuracy and efficiency.

[0095] In one embodiment, such as Figure 2 As shown, the step of extracting knife mark texture features based on target wire core data and constructing a deep learning-based denoising encoding / decoding network model based on the knife mark texture features includes:

[0096] Step S201: Obtain a cross-sectional diagram of the wire core with knife-mark texture based on the target wire core data, and obtain the conductor structure diagram of the cable core from the database;

[0097] Step S202: Extract the image features of the wire core cross-section diagram and the cable wire core conductor structure diagram respectively, compare them to determine the difference image features, and determine the knife mark texture description features based on the difference image features;

[0098] Step S203: Determine the mapping nonlinear parameters of the knife mark texture image and the removed knife mark texture image based on the knife mark texture description features;

[0099] Step S204: Select a deep learning-based autoencoder denoising model architecture and construct a denoising encoder-decoder network model by using the mapping nonlinear parameters of the knife mark texture image and the image with the knife mark removed.

[0100] The beneficial effects of the above technical solution are as follows: by determining the mapping nonlinear parameters of the knife mark texture image and the knife mark texture image, the denoising encoding and decoding network model can be constructed to ensure that the working mechanism of the model meets the requirements.

[0101] In one embodiment, training a deep learning-based denoising codec network model using a wire core dataset and verifying the model's effectiveness in eliminating knife mark textures includes:

[0102] Multiple original wire core images in the wire core dataset are rotated to obtain multiple incremental wire core images. The incremental wire core images and the original wire core images are integrated and proportionally allocated into the model training set, the model test set, and the model validation set.

[0103] In the model training set, a preset batch of wire core images are randomly selected to train the deep learning-based denoising encoding and decoding network model. The model output image is collected and compared with the image with knife mark texture to determine the model loss.

[0104] The network model parameters are optimized based on the model loss until the model converges. The converged denoising codec network model is tested using a model test set and validated using a model validation set.

[0105] The success rate of the denoising codec network model in removing knife marks was determined based on the validation data, and the effectiveness of the denoising codec network model in removing knife marks was determined based on the success rate.

[0106] The beneficial effects of the above technical solution are: it can effectively test and verify the model to ensure the elimination effect of knife mark texture, and improve the stability and reliability of the model.

[0107] In one embodiment, the step of determining model validity and accuracy based on verification results and performing a model qualification judgment, and if the judgment is qualified, using a denoising encoding / decoding network model to remove subsequent knife-marked wire core images, includes:

[0108] Based on the verification results, the correct and incorrect segmentation sample data of the denoising codec network model after the removal of knife mark texture were determined.

[0109] The effectiveness and accuracy of the denoising encoder-decoder network model are determined based on correctly segmented and incorrectly segmented sample data.

[0110] Based on preset effectiveness and accuracy indicators, the denoising codec network model is judged to be qualified according to the model effectiveness and model accuracy.

[0111] If the result is deemed satisfactory, subsequent images of wire cores with knife marks are acquired and preprocessed. A denoising codec network model is then used to remove the knife mark texture from the preprocessed images of the wire cores with knife marks, and an output image is obtained.

[0112] The beneficial effects of the above technical solution are as follows: by determining the effectiveness and accuracy of the model based on the proportion of sample data, the working accuracy and qualification of the model can be objectively and reasonably judged based on the output results of the model, thereby using the model to process subsequent core images, which improves practicality and stability.

[0113] In one embodiment, determining the correctly segmented and incorrectly segmented sample data of the denoising codec network model for the knife mark texture removal based on the verification results includes:

[0114] Based on the image features of samples before and after knife mark texture removal, a watershed segmentation threshold is set, and multiple model output images are obtained based on the verification results.

[0115] For each model output image, extract the core contour and calculate the current number of cores and the area of ​​a single conductor within the extracted contour area;

[0116] The segmented conductor contour edge parameters corresponding to each model output image are determined based on the current number of conductors within the contour region and the area of ​​a single conductor.

[0117] The current watershed segmentation value of each model output image is determined based on the edge parameters of the segmented conductor contour. The current watershed segmentation value is compared with the watershed segmentation threshold. The first model output image with the same value is taken as the correctly segmented sample data, and the second model output image with different value is taken as the incorrectly segmented sample data.

[0118] The beneficial effects of the above technical solution are as follows: by setting the watershed segmentation threshold, the edge contour segmentation parameters can be determined by the best reference sample, thereby making reference judgment and sample classification on the watershed segmentation value of the real-time output image of the model, which improves the classification efficiency and reliability.

[0119] In one embodiment, this embodiment also discloses a deep learning-based wire core tool mark texture removal and detection system, such as... Figure 3 As shown, the system includes:

[0120] Extraction module 301 is used to construct a machine vision-based wire core recognition and detection component and collect a wire core dataset through it, and extract target wire core data containing knife mark texture from the wire core dataset;

[0121] Module 302 is used to extract knife mark texture features based on target wire core data and construct a deep learning-based denoising encoding / decoding network model based on the knife mark texture features.

[0122] The verification module 303 is used to train a deep learning-based denoising codec network model using a wire core dataset and to verify the model's effectiveness in eliminating knife mark textures.

[0123] The elimination module 304 is used to determine the effectiveness and accuracy of the model based on the verification results and to make a model qualification judgment. If the judgment is qualified, the noise reduction codec network model is used to eliminate the subsequent knife-marked wire core images.

[0124] The working principle and beneficial effects of the above technical solution have been explained in the method embodiments, and will not be repeated here.

[0125] In one embodiment, the extraction module 301 includes:

[0126] The configuration submodule is used to obtain the reflection parameters of the wire core, determine the lens refraction angle based on the reflection parameters, and configure the relative position parameters of the coaxial light source and the diffuse reflector according to the lens refraction angle.

[0127] The first determination submodule is used to determine the edge structure parameters of the cable core, determine the pixel error based on the edge structure parameters, and determine the resolution requirement based on the pixel error.

[0128] The generation submodule is used to select pixel configuration parameters based on resolution requirements, select a target specification monochrome CCD camera based on the pixel configuration parameters, and generate a machine vision-based core recognition and detection component through the target specification monochrome CCD camera, coaxial light source, diffuse reflector, and lens.

[0129] The second determination submodule is used to acquire multiple qualified cable core images through the core identification and detection component and generate a core dataset to determine the identification parameters of the knife mark texture.

[0130] The extraction submodule is used to extract target cable core images containing knife mark textures from the core dataset based on recognition parameters and generate target core data.

[0131] In one embodiment, such as Figure 4 As shown, the construction module 302 includes:

[0132] The acquisition submodule 3021 is used to obtain a cross-sectional diagram of the wire core with knife mark texture based on the target wire core data, and to obtain a conductor structure diagram of the cable core from the database;

[0133] The third determination submodule 3022 is used to extract the image features of the wire core cross-section diagram and the cable wire core conductor structure diagram respectively, compare and determine the difference image features, and determine the knife mark texture description features based on the difference image features.

[0134] The fourth determining submodule 3023 is used to determine the mapping nonlinear parameters of the knife mark texture image and the removed knife mark texture image based on the knife mark texture description features;

[0135] Submodule 3024 is constructed to select the deep learning-based autoencoder denoising model architecture and construct the denoising encoder-decoder network model by mapping nonlinear parameters of the knife mark texture image and the knife mark texture image removed.

[0136] In one embodiment, the verification module 303 includes:

[0137] The allocation submodule is used to rotate multiple original core images in the core dataset to obtain multiple incremental core images, integrate the incremental core images and original core images, and allocate them proportionally as a model training set, a model test set, and a model validation set.

[0138] The fifth determination submodule is used to randomly select a preset batch of wire core images in the model training set to train the deep learning-based denoising encoding and decoding network model, collect the model output image and compare it with the knife mark texture image to determine the model loss.

[0139] The validation submodule is used to optimize the network model parameters based on the model loss until the model converges, test the converged denoising codec network model with the model test set, and validate the converged denoising codec network model with the model validation set.

[0140] The sixth determination submodule is used to determine the success rate of the denoising codec network model in removing knife marks based on the verification data, and to determine the effect of the denoising codec network model on removing knife marks based on the success rate.

[0141] The elimination module 304 includes:

[0142] The seventh determination submodule is used to determine the correct and incorrect segmentation sample data of the denoising codec network model after the removal of knife marks and textures, based on the verification results.

[0143] The eighth determination submodule is used to determine the model validity and model accuracy of the denoising encoder-decoder network model based on correctly segmented sample data and incorrectly segmented sample data;

[0144] The judgment submodule is used to determine the pass / fail status of the denoising codec network model based on the model's validity and accuracy according to preset validity and accuracy indicators.

[0145] The elimination submodule is used to acquire subsequent images of wire cores with knife marks if the judgment is qualified, and preprocess them. Then, the knife mark texture is eliminated by the preprocessed subsequent images of wire cores with knife marks through a denoising codec network model to obtain the output image.

[0146] Those skilled in the art should understand that the "first" and "second" in this invention simply refer to different application stages.

[0147] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0148] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for eliminating tool marks and textures on wire cores based on deep learning, characterized in that, Includes the following steps: A machine vision-based core identification and detection component is constructed and a core dataset is collected using it. Target core data containing knife mark textures is then extracted from the core dataset. Based on the target wire core data, knife mark texture features are extracted, and a deep learning-based denoising encoding and decoding network model is constructed based on the knife mark texture features; A deep learning-based denoising codec network model was trained using a wire core dataset, and the model's effectiveness in eliminating knife mark textures was verified. The validity and accuracy of the model are determined based on the verification results, and the model is deemed qualified. If the model is deemed qualified, the noise reduction encoding and decoding network model is used to remove the knife-marked wire core images in the subsequent process.

2. The method for eliminating wire core tool mark texture based on deep learning according to claim 1, characterized in that, The construction of a machine vision-based wire core recognition and detection component, and its acquisition of a wire core dataset, includes extracting target wire core data containing knife mark textures from the dataset. Obtain the reflection parameters of the wire core, determine the lens refraction angle based on the reflection parameters, and configure the relative position parameters of the coaxial light source and the diffuse reflector according to the lens refraction angle; Determine the edge structure parameters of the cable core, determine the pixel error based on the edge structure parameters, and determine the resolution requirement based on the pixel error. Pixel configuration parameters are selected based on resolution requirements. A target specification monochrome CCD camera is selected based on the pixel configuration parameters. A machine vision-based core recognition and detection component is generated by using the target specification monochrome CCD camera, a coaxial light source, a diffuse reflector, and a lens. Multiple qualified cable core images are acquired by the core identification and detection component and a core dataset is generated to determine the identification parameters of knife mark texture. Based on the recognition parameters, target cable core images containing knife mark textures are extracted from the core dataset and target core data is generated.

3. The method for eliminating wire core tool mark texture based on deep learning according to claim 1, characterized in that, The step of extracting knife mark texture features based on target wire core data and constructing a deep learning-based denoising encoding / decoding network model based on the knife mark texture features includes: Obtain a cross-sectional diagram of the wire core with knife-mark texture based on the target wire core data, and retrieve the conductor structure diagram of the cable core from the database; The image features of the cross-sectional image of the wire core and the structural image of the cable core are extracted and compared to determine the differences in image features. Based on the differences in image features, the texture description features of the knife marks are determined. The mapping nonlinear parameters between the knife mark texture image and the removed knife mark texture image are determined based on the knife mark texture description features; The architecture of a deep learning-based autoencoder denoising model is selected by mapping nonlinear parameters between the knife-mark texture image and the image with the knife-mark removed, and a denoising encoder-decoder network model is constructed.

4. The method for eliminating wire core tool mark texture based on deep learning according to claim 1, characterized in that, The step of training a deep learning-based denoising encoding / decoding network model using a wire core dataset and verifying the model's effectiveness in eliminating knife mark textures includes: Multiple original wire core images in the wire core dataset are rotated to obtain multiple incremental wire core images. The incremental wire core images and the original wire core images are integrated and proportionally allocated into the model training set, the model test set, and the model validation set. In the model training set, a preset batch of wire core images are randomly selected to train the deep learning-based denoising encoding and decoding network model. The model output image is collected and compared with the image with knife mark texture to determine the model loss. The network model parameters are optimized based on the model loss until the model converges. The converged denoising codec network model is tested using a model test set and validated using a model validation set. The success rate of the denoising codec network model in removing knife marks was determined based on the validation data, and the effectiveness of the denoising codec network model in removing knife marks was determined based on the success rate.

5. The method for eliminating wire core tool mark texture based on deep learning according to claim 1, characterized in that, The process involves determining the model's effectiveness and accuracy based on the verification results and then judging its suitability. If the model is deemed suitable, a denoising encoding / decoding network model is used to remove the knife-marked wire core images in subsequent processing, including: Based on the verification results, the correct and incorrect segmentation sample data of the denoising codec network model after the removal of knife mark texture were determined. The effectiveness and accuracy of the denoising encoder-decoder network model are determined based on correctly segmented and incorrectly segmented sample data. Based on preset effectiveness and accuracy indicators, the denoising codec network model is judged to be qualified according to the model effectiveness and model accuracy. If the result is deemed satisfactory, subsequent images of wire cores with knife marks are acquired and preprocessed. A denoising codec network model is then used to remove the knife mark texture from the preprocessed images of the wire cores with knife marks, and an output image is obtained.

6. The method for eliminating wire core tool mark texture based on deep learning according to claim 5, characterized in that, The step of determining the correct and incorrect segmentation sample data of the denoising encoding / decoding network model for the knife mark texture removal based on the verification results includes: Based on the image features of samples before and after knife mark texture removal, a watershed segmentation threshold is set, and multiple model output images are obtained based on the verification results. For each model output image, extract the core contour and calculate the current number of cores and the area of ​​a single conductor within the extracted contour area; The segmented conductor contour edge parameters corresponding to each model output image are determined based on the current number of conductors within the contour region and the area of ​​a single conductor. The current watershed segmentation value of each model output image is determined based on the edge parameters of the segmented conductor contour. The current watershed segmentation value is compared with the watershed segmentation threshold. The first model output image with the same value is taken as the correctly segmented sample data, and the second model output image with different value is taken as the incorrectly segmented sample data.

7. A deep learning-based system for detecting and eliminating tool marks and textures on wire cores, characterized in that, The system includes: The extraction module is used to build a machine vision-based wire core recognition and detection component and collect a wire core dataset to extract target wire core data containing knife mark texture from the wire core dataset. The module is used to extract knife mark texture features based on target wire core data and construct a deep learning-based denoising encoding and decoding network model based on the knife mark texture features. The verification module is used to train a deep learning-based denoising codec network model using a wire core dataset and to verify the model's effectiveness in eliminating knife mark textures. The elimination module is used to determine the effectiveness and accuracy of the model based on the verification results and to make a model qualification judgment. If the judgment is qualified, the noise reduction codec network model is used to eliminate the subsequent knife-marked wire core images.

8. The deep learning-based wire core tool mark texture removal and detection system according to claim 7, characterized in that, The extraction module includes: The configuration submodule is used to obtain the reflection parameters of the wire core, determine the lens refraction angle based on the reflection parameters, and configure the relative position parameters of the coaxial light source and the diffuse reflector according to the lens refraction angle. The first determination submodule is used to determine the edge structure parameters of the cable core, determine the pixel error based on the edge structure parameters, and determine the resolution requirement based on the pixel error. The generation submodule is used to select pixel configuration parameters based on resolution requirements, select a target specification monochrome CCD camera based on the pixel configuration parameters, and generate a machine vision-based core recognition and detection component through the target specification monochrome CCD camera, coaxial light source, diffuse reflector, and lens. The second determination submodule is used to acquire multiple qualified cable core images through the core identification and detection component and generate a core dataset to determine the identification parameters of the knife mark texture. The extraction submodule is used to extract target cable core images containing knife mark textures from the core dataset based on recognition parameters and generate target core data.

9. The deep learning-based wire core tool mark texture removal and detection system according to claim 7, characterized in that, The building module includes: The acquisition submodule is used to obtain a cross-sectional diagram of the wire core with knife-mark texture based on the target wire core data, and to obtain the conductor structure diagram of the cable core from the database; The third determination submodule is used to extract the image features of the wire core cross-section diagram and the cable wire core conductor structure diagram respectively, compare them to determine the difference image features, and determine the knife mark texture description features based on the difference image features. The fourth determination submodule is used to determine the mapping nonlinear parameters of the knife mark texture image and the removed knife mark texture image based on the knife mark texture description features; A submodule is constructed to select a deep learning-based autoencoder denoising model architecture and construct a denoising encoder-decoder network model by mapping nonlinear parameters between the knife-mark texture image and the image with the knife-mark texture removed.

10. The deep learning-based wire core tool mark texture removal and detection system according to claim 7, characterized in that, The verification module includes: The allocation submodule is used to rotate multiple original core images in the core dataset to obtain multiple incremental core images, integrate the incremental core images and original core images, and allocate them proportionally as a model training set, a model test set, and a model validation set. The fifth determination submodule is used to randomly select a preset batch of wire core images in the model training set to train the deep learning-based denoising encoding and decoding network model, collect the model output image and compare it with the knife mark texture image to determine the model loss. The validation submodule is used to optimize the network model parameters based on the model loss until the model converges, test the converged denoising codec network model with the model test set, and validate the converged denoising codec network model with the model validation set. The sixth determination submodule is used to determine the success rate of the denoising codec network model in removing knife marks based on the verification data, and to determine the effect of the denoising codec network model on removing knife marks based on the success rate. The elimination module includes: The seventh determination submodule is used to determine the correct and incorrect segmentation sample data of the denoising codec network model after the removal of knife marks and textures, based on the verification results. The eighth determination submodule is used to determine the model validity and model accuracy of the denoising encoder-decoder network model based on correctly segmented sample data and incorrectly segmented sample data; The judgment submodule is used to determine the pass / fail status of the denoising codec network model based on the model's validity and accuracy according to preset validity and accuracy indicators. The elimination submodule is used to acquire subsequent images of wire cores with knife marks if the judgment is qualified, and preprocess them. Then, the knife mark texture is eliminated by the preprocessed subsequent images of wire cores with knife marks through a denoising codec network model to obtain the output image.

Images