High-speed copper cable surface quality evaluation method and system based on machine vision

Abstract

The invention relates to the field of copper cables, in particular to a high-speed copper cable surface quality evaluation method and system based on machine vision. Comprising the following steps: constructing an image acquisition system to obtain image data of the high-speed copper cable; preprocessing the image data; establishing a surface defect classification model based on deep learning, inputting the preprocessed image data, and outputting a defect judgment result and a defect quantized value of the image data; and calculating a defect area proportion and a defect depth estimation value based on the preprocessed image data and a defect judgment result, inputting the defect area proportion and the defect depth estimation value into a comprehensive judgment model to calculate a defect severity evaluation value, and generating a surface quality evaluation report of the high-speed copper cable based on the defect severity evaluation value. According to the invention, an image acquisition system composed of a plurality of high-speed cameras and a uniform annular light source is established, so that images can be clearly captured in the high-speed copper cable production process, and the image acquisition quality is ensured.

Description

Technical Field

[0001] This invention relates to the field of copper cables, and more particularly to a method and system for evaluating the surface quality of high-speed copper cables based on machine vision. Background Technology

[0002] In the copper cable manufacturing industry, the surface quality of high-speed copper cables directly affects their electrical performance and mechanical strength, making accurate surface quality assessment crucial. Traditional methods for assessing copper cable surface quality primarily rely on manual visual inspection, which has significant drawbacks. On one hand, manual inspection is inefficient and cannot meet the real-time inspection requirements of high-speed production lines; on the other hand, the inspection results are greatly influenced by the subjective factors of the inspectors, easily leading to missed detections or misjudgments.

[0003] With the development of machine vision technology, some machine vision-based inspection methods have been applied to the surface quality assessment of copper cables. However, existing technologies still face a key technical problem when dealing with surface defect detection in high-speed copper cables: it is difficult to accurately identify minute defects on the surface of high-speed copper cables during the production process, and it is impossible to comprehensively and quantitatively assess the severity of defects. Specifically, existing methods do not capture clear images of high-speed copper cables during the production process, have insufficient feature extraction capabilities, especially for the identification of minute defects, and lack comprehensive consideration of key parameters, making it difficult to provide scientific and quantitative quality assessment results.

[0004] Therefore, it is necessary to develop a machine vision-based method and system for evaluating the surface quality of high-speed copper cables. Summary of the Invention

[0005] To address the technical deficiencies in the background technology, this invention proposes a machine vision-based method and system for evaluating the surface quality of high-speed copper cables, solving the aforementioned technical problems and meeting practical needs. The specific technical solution is as follows: A machine vision-based method for assessing the surface quality of high-speed copper cables includes the following steps: An image acquisition system is built by using a camera that surrounds a high-speed copper cable and a uniform ring light source to acquire image data of the high-speed copper cable. The image data is preprocessed, which includes image grayscale conversion, filtering and noise reduction, edge detection and morphological processing in sequence. Establish a surface defect classification model based on deep learning, input preprocessed image data, and output the defect judgment result and defect quantification value of the image data; Based on the preprocessed image data and defect judgment results, the defect area ratio and defect depth estimate are calculated. The defect quantification value, defect area ratio and defect depth estimate are input into the comprehensive judgment model to calculate the defect severity assessment value. Based on the defect severity assessment value, a surface quality assessment report of high-speed copper cable is generated.

[0006] Furthermore, the image grayscale conversion specifically includes: converting the color image of the image data into a grayscale image; the filtering and denoising specifically includes: applying a Gaussian filtering algorithm to denoise the grayscale image; the edge detection specifically includes: using an edge detection algorithm to extract edges from the denoised image to obtain the image edge information of the high-speed copper cable; the morphological processing specifically includes: using dilation and erosion operations to optimize the edge-extracted image; wherein, the Gaussian kernel size in the Gaussian filtering algorithm is adaptively adjusted according to the surface texture complexity of the high-speed copper cable.

[0007] Furthermore, the surface defect classification model adopts an improved convolutional neural network structure with an attention mechanism module. When establishing the surface defect classification model, the model is trained using high-speed copper cable image data containing different types of defects. The learning rate during model training adopts a dynamic adjustment strategy, which specifically involves setting an initial learning rate and gradually decreasing it as the number of training rounds increases.

[0008] Furthermore, when establishing the surface defect classification model, the specific working process of the attention mechanism module is as follows: Obtain a feature map of the input image data, the feature map having three dimensions: number of channels, height, and width; Global average pooling is performed on the feature map to compress it in the height and width dimensions, resulting in a one-dimensional vector that retains only the channel dimension information. The one-dimensional vector is sequentially input into two fully connected layers, and a non-linear activation function is used between the two fully connected layers to process it, thereby obtaining the attention weight vector, the dimension of which is consistent with the number of channels; The attention weight vector is weighted with the input feature map to enhance the feature information related to defects in the feature map, and the weighted feature map is output.

[0009] Furthermore, the defect area ratio is obtained by calculating the ratio of the number of pixels in the defect area to the total number of pixels in the image. The defect depth estimate is calculated using a method based on image grayscale changes. By analyzing the grayscale difference between the defect area and the surrounding normal area, and combining it with the optical property parameters of the copper cable surface material, the following formula is used for calculation:

[0010] Where D is the estimated defect depth, and k is a coefficient related to high-speed copper cables. This represents the grayscale difference between the defective area and the surrounding normal area. This represents the average grayscale value of the normal area. The formula for calculating the percentage of defect area is as follows:

[0011] in, This represents the percentage of the defective area. This represents the number of pixels in the defect area. This represents the total number of pixels in the entire image.

[0012] Furthermore, the formula for calculating the severity assessment value of defects in the comprehensive judgment model is as follows:

[0013] in, This is a value used to assess the severity of the defect. For defect quantification values, corresponding quantification values ​​are pre-set for different types of defects. This represents the percentage of the defective area. This is an estimate of the defect depth. , and These are the weighting coefficients for the defect quantification value, defect area ratio, and defect depth estimate, respectively, and they satisfy the following conditions: The weighting coefficient , and The settings are determined based on the specifications, application scenarios, and industry standards of high-speed copper cables, and the optimal values ​​are determined through experiments and data analysis.

[0014] Furthermore, the specific steps for generating a surface quality assessment report for high-speed copper cables based on defect severity assessment values ​​include: The severity assessment value of the defect is compared with a pre-set first threshold, second threshold, and third threshold to obtain the severity level of the defect. Collect key data generated throughout the evaluation process, including the acquisition parameters of the image acquisition system, defect determination results, defect quantification values, defect area ratio, defect depth estimation values, and defect severity assessment values. Based on the severity level of defects and key data, corresponding processing suggestions are obtained from a pre-set defect processing knowledge base. The severity level of defects, key data and processing suggestions are then summarized into a surface quality assessment report.

[0015] Furthermore, after generating the surface quality assessment report for the high-speed copper cable based on the defect severity assessment value, the process also includes the following steps: Store the surface quality assessment report for each evaluation in the database; Regularly perform statistical analysis on the surface quality assessment reports in the database to generate trend reports on the surface quality of high-speed copper cables. The statistical analysis process uses time series analysis methods to predict the frequency and severity of defects in different time periods through a preset prediction model. The parameters of the prediction model are determined by the maximum likelihood estimation method, and the prediction accuracy of the prediction model is evaluated by the root mean square error. The formula for calculating the root mean square error is:

[0016] in, Let n be the root mean square error, and n be the number of samples. This is the actual value. These are predicted values.

[0017] Furthermore, the specific determination logic for the severity level of the defect is as follows: When the defect severity assessment value is 0, it is determined that there is no defect. When the defect severity assessment value is greater than 0 and less than or equal to the first threshold, it is determined that there is a minor defect. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to strengthen monitoring in subsequent production processes. When the defect severity assessment value is greater than the first threshold and less than or equal to the second threshold, it is determined that there is a general defect. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to adjust the production process parameters. When the defect severity assessment value is greater than the second threshold and less than or equal to the third threshold, a serious defect is determined to exist. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to stop production immediately and inspect and repair the relevant equipment. When the defect severity assessment value is greater than the third threshold, it is determined that there is an extremely serious defect. At this time, the handling suggestions in the defect handling knowledge base include: in addition to stopping production and inspecting the equipment, a comprehensive review of the copper cables that have been produced is also required. The defect handling knowledge base was established through the analysis and summarization of a large number of historical defect cases.

[0018] A machine vision-based surface quality assessment system for high-speed copper cables includes: The image acquisition module is used to acquire images of the surface of high-speed copper cables. It includes an image acquisition unit composed of several high-speed cameras and a uniform ring light source. The image preprocessing module is electrically connected to the image acquisition module. It is used to receive the acquired image data and perform preprocessing. It is equipped with a grayscale conversion unit, a filtering and noise reduction unit, an edge detection unit, and a morphological processing unit. The defect classification module is equipped with a defect classification model, which is used to receive preprocessed image data, classify and judge defects, and output defect judgment results and defect quantification values. The defect assessment module is equipped with a comprehensive judgment model, which calculates the severity assessment value of the defect based on the quantitative value of the defect judgment result, the defect area ratio, and the defect depth estimate. This allows for the determination of the severity of the surface defects of the copper cable and the assessment of the surface quality of the copper cable, generating an assessment report. The defect assessment module has built-in weighting coefficients set according to the specifications of the copper cable, application scenarios, and industry standards, as well as thresholds determined according to the enterprise's quality control requirements and production experience. The data storage module is equipped with an analysis unit to store evaluation reports and perform periodic statistical analysis to generate copper cable surface quality trend reports.

[0019] Compared with existing technologies, the machine vision-based method and system for evaluating the surface quality of high-speed copper cables provided by this invention have the following advantages: This invention constructs an image acquisition system composed of multiple high-speed cameras and a uniform ring light source, enabling clear image capture during high-speed copper cable production and ensuring high-quality image acquisition. Simultaneously, by employing an improved convolutional neural network structure and adding an attention mechanism module to the surface defect classification model, combined with a training strategy that dynamically adjusts the learning rate, the model's attention to and recognition ability of subtle defects on the copper cable surface is significantly improved. This achieves accurate identification of minute defects during the production process, significantly enhancing defect recognition accuracy and detection efficiency.

[0020] This invention introduces defect area ratio and defect depth estimation, and uses a comprehensive judgment model to weight and calculate the defect severity assessment value by weighting the defect quantification value, defect area ratio, and defect depth estimation value. This multi-parameter comprehensive quantitative assessment method changes the limitation of existing technologies that rely solely on defect type for assessment, and can comprehensively and scientifically reflect the degree of impact of defects on the surface quality of copper cables. By comparing the assessment value with preset thresholds, different severity levels are divided, making the assessment results clearer and more intuitive, and providing an accurate basis for production decisions. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the machine vision-based method for evaluating the surface quality of high-speed copper cables in this invention.

[0022] Figure 2This is a schematic diagram of the modules of the machine vision-based high-speed copper cable surface quality assessment system of the present invention. Detailed Implementation In the description of this invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "middle," and "inner," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, it should be noted that unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention through specific circumstances.

[0023] The embodiments of the present invention will be described below with reference to the accompanying drawings and related examples. The embodiments of the present invention are not limited to the following examples, and the present invention relates to the relevant necessary components in this technical field, which should be regarded as well-known technology in this technical field and can be known and mastered by those skilled in this technical field.

[0024] See Figure 1 This invention provides a machine vision-based method for evaluating the surface quality of high-speed copper cables, comprising the following steps: Step S100: An image acquisition system is built by using a camera and a uniform ring light source that surrounds the high-speed copper cable to acquire image data of the high-speed copper cable.

[0025] The acquisition camera system, which surrounds the high-speed copper cable, comprises multiple high-speed cameras arranged in a ring around the cable. These cameras are fixed by mechanical supports, and their lens angles are adjusted to cover the cable surface. The high-speed cameras include industrial cameras with high frame rates and high resolution. The uniform ring light source consists of a ring-shaped illumination device surrounding the copper cable, composed of multiple evenly distributed LED light sources. Illumination uniformity is optimized by adjusting the brightness and angle of the light sources. By synchronously installing the camera and light source alongside the production line, with the camera lens facing the cable surface and synchronized with the production line speed via a trigger signal, clear images are continuously captured. This operation eliminates shadows and reflections, solving the image quality problems caused by motion blur or uneven illumination in traditional methods, providing high-quality raw data for subsequent processing.

[0026] Step S200: Preprocess the image data, which includes image grayscale conversion, filtering and noise reduction, edge detection and morphological processing in sequence.

[0027] Step S300: Establish a surface defect classification model based on deep learning, input preprocessed image data, and output the defect judgment result and defect quantification value of the image data.

[0028] Step S400: Based on the preprocessed image data and defect judgment results, calculate the defect area ratio and defect depth estimate. Input the defect quantification value, defect area ratio and defect depth estimate into the comprehensive judgment model to calculate the defect severity assessment value. Generate a surface quality assessment report for high-speed copper cables based on the defect severity assessment value.

[0029] The defect area ratio is a quantitative indicator calculated by binarization segmentation, representing the ratio of the defect area to the overall image area, achieved through threshold segmentation or connected component analysis. The defect depth estimate is a parameter derived from the image's gray-level gradient, calculated through structured light scanning or gray-level change rate calculation. The comprehensive judgment model is a mathematical model that integrates multi-dimensional indicators using a weighted algorithm, including an evaluation function of linear weighted summation. Step S400 involves first locating the defect area based on the classification results, calculating the area ratio through threshold segmentation, calculating the depth estimate by combining gray-level gradient prediction, and finally inputting all three into a weighted model to generate a comprehensive evaluation value, determining the defect level based on the threshold. This operation, through the synergistic analysis of multi-dimensional quantitative indicators, overcomes the limitations of traditional methods that rely on a single defect type for evaluation, achieving a scientific quantification of defect severity.

[0030] This invention achieves the technical effects of improving detection efficiency and accuracy, reducing the need for manual intervention, and adapting to real-time production line environments by optimizing the image acquisition system to obtain high-quality raw data, progressively enhancing defect visibility through preprocessing steps, realizing high-precision classification through an improved deep learning model, and quantifying defect severity through a multi-parameter comprehensive model. Through full-process automation and multi-dimensional quantitative evaluation, this solution significantly reduces the inefficiencies, insufficient accuracy, and subjective judgment errors of traditional methods, providing intelligent decision support for copper cable quality control.

[0031] In one embodiment of the present invention, the image grayscale conversion specifically includes: converting a color image of image data into a grayscale image; This step achieves dimensionality reduction of the original image by converting multi-channel color image data into a single-channel grayscale image. The grayscale image can be generated using a weighted average method, the purpose of which is to simplify the amount of data for subsequent processing and preserve key texture features.

[0032] The filtering and denoising specifically includes: using a Gaussian filtering algorithm to denoise the grayscale image; The Gaussian filtering algorithm uses a two-dimensional normal distribution matrix to perform convolution operations on the image to eliminate random noise. The size parameter of the Gaussian kernel is dynamically adjusted according to the surface texture complexity of the copper cable.

[0033] The edge detection specifically includes: using an edge detection algorithm to extract the edges of the denoised image to obtain the image edge information of the high-speed copper cable; The edge detection algorithm identifies potential defect boundaries by analyzing regions of abrupt changes in image grayscale. This step relies on high-quality, filtered image input to reduce false detections caused by noise interference. Images processed with adaptive Gaussian filtering can more clearly present defect edge features, thereby improving the localization accuracy of the edge detection algorithm.

[0034] The morphological processing specifically includes: optimizing the image after edge extraction using dilation and erosion operations; The morphological processing involves transforming the binarized edge image using structured elements. Dilation connects broken edges and fills internal pores in defects, while erosion removes edge burrs or isolated noise points. This step further refines the defect contour based on the optimized edge information, providing more accurate geometric features for subsequent classification.

[0035] The size of the Gaussian kernel in the Gaussian filtering algorithm is adaptively adjusted according to the surface texture complexity of the high-speed copper cable.

[0036] The texture complexity is quantified by calculating the standard deviation of pixel grayscale values ​​in local regions, and the grayscale variance within the image region is analyzed using a sliding window. When a high-complexity region is detected, the system selects a larger Gaussian kernel to adequately suppress noise; while for low-complexity regions, a smaller kernel is used to reduce detail blurring. This dynamic adjustment mechanism is implemented through preset rules or trained models, such as determining the kernel size based on gradient change thresholds or texture feature classification results. This operation solves the problem of traditional fixed-kernel Gaussian filtering over-smoothing in complex texture regions or insufficient denoising in simple regions, ensuring a balance between noise suppression and detail preservation.

[0037] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. By converting color images to grayscale to reduce data dimensionality, an adaptive Gaussian filtering algorithm dynamically adjusts the kernel size based on texture complexity for targeted denoising. This is combined with an edge detection algorithm to extract defect boundary features, and morphological operations are used to optimize edge contours. This achieves a dynamic balance between denoising and texture preservation, resulting in more accurate edge detection and more complete defect contours after morphological processing. This method effectively overcomes the limitations of traditional fixed-parameter filtering in complex texture scenarios, significantly improves the compatibility of noise suppression and detail preservation, and enhances the ability of subsequent edge detection and classification models to identify subtle defects. Ultimately, it enables highly robust assessment of copper cable surface quality in high-speed production lines.

[0038] In one embodiment of the present invention, the surface defect classification model adopts an improved convolutional neural network structure and has an attention mechanism module. When establishing the surface defect classification model, the model is trained using high-speed copper cable image data containing different types of defects. The learning rate during model training is dynamically adjusted by sampling. Specifically, the dynamic adjustment strategy is to set an initial learning rate and gradually decrease it as the number of training rounds increases.

[0039] The improved convolutional neural network structure is a deep learning framework optimized by introducing an attention mechanism module, constructed by stacking convolutional layers, pooling layers, and nonlinear activation functions. The attention mechanism module is a computational unit that strengthens the feature weights of defective regions through channel attention, implemented through a combination of global average pooling and fully connected layers. The dynamic adjustment strategy is a control method for adjusting the learning rate parameter according to changes in training epochs, achieved through a preset decay function, including exponential decay, linear decay, or cosine annealing strategies. The initial learning rate is a baseline value for parameter updates in the early stages of training, obtained through experimental verification or empirical setting, such as 0.001. The learning rate decay rule defines a specific mathematical function for how the learning rate changes with each epoch, achieved by multiplying by a decay factor after each fixed epoch or calculating the current value based on a cosine function, such as multiplying by a decay factor of 0.9 every 5 epochs.

[0040] The dynamic adjustment strategy can be implemented by embedding a decay function in the training loop to calculate the current learning rate. Specifically, this involves: first, setting an initial learning rate value and defining a decay rule; then, at the beginning of each round, calculating the updated learning rate based on the current round, and applying this value to the gradient update step in backpropagation. For example, when using a cosine annealing strategy, the learning rate calculation formula can be expressed as the initial learning rate multiplied by (1 + cos(π × current round / total rounds)) / 2. This operation controls the parameter update step size in stages, allowing the model to quickly adapt to the data distribution characteristics in the early stages of training, and gradually focus on optimizing the classification boundaries of subtle defects in later stages, thus balancing convergence speed and parameter optimization accuracy. Specifically, a high initial learning rate allows the model to quickly capture global features of defects, such as edge distribution, in the first few rounds, while a lower decaying learning rate reduces the magnitude of parameter updates, allowing the attention mechanism module to more stably strengthen the feature weight allocation of key regions, ultimately improving the recognition accuracy of subtle defects such as 0.1mm-level micro-dents.

[0041] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. It employs an improved convolutional neural network structure combined with an attention mechanism to enhance defect region features. The method is trained using an image dataset containing multiple types of defects, and a dynamic learning rate adjustment strategy that decreases with each training round is implemented. The learning rate parameter is calculated in real-time according to the current training round based on a preset decay rule, significantly improving model training efficiency and classification accuracy. This method accelerates the model's adaptation to global features with a high initial learning rate, and then reduces parameter oscillations by gradually decreasing the learning rate to optimize the ability to discriminate subtle features. Simultaneously, the attention mechanism strengthens key regions, shortening the training cycle and enhancing robustness to complex defects caused by blurring or changes in lighting conditions in high-speed production environments. Ultimately, it provides more reliable classification results for defect quantification assessment.

[0042] In one embodiment of the present invention, when establishing a surface defect classification model, the specific working process of the attention mechanism module is as follows: A feature map of the input image data is acquired, the feature map having three dimensions: number of channels, height, and width; a global average pooling operation is performed on the feature map to compress it in the height and width dimensions, resulting in a one-dimensional vector that retains only channel dimension information; this one-dimensional vector is sequentially input into two fully connected layers, and processed using a non-linear activation function between the two fully connected layers to obtain an attention weight vector, the dimension of which is consistent with the number of channels; the attention weight vector is weighted with the input feature map to enhance the defect-related feature information in the feature map, and the weighted feature map is output.

[0043] The feature map is a multidimensional tensor extracted from the convolutional neural network. Its three-dimensional structure consists of the number of channels, height, and width, and it is obtained through the feature extraction process of the preceding convolutional layers. The feature map includes a 64×7×7 tensor output by the convolutional layer. Global average pooling is a feature compression technique that calculates the average value across the spatial dimensions of each channel. By averaging the height and width dimensions of the feature map, for example, when the input feature map size is 64×7×7, global average pooling can yield a 64×1×1 one-dimensional vector. The fully connected layer is a combination structure of linear transformation and nonlinear activation composed of multiple neuron nodes. It linearly maps the input vector through a weight matrix. For example, the first fully connected layer can compress a 64-dimensional vector to 16 dimensions, and the second fully connected layer can restore it to 64 dimensions. The nonlinear activation function is a mathematical function that introduces nonlinear characteristics, including the ReLU function. The ReLU function can preserve positive response features through the max(0,x) operation. The attention weight vector is a scalar sequence with the same number of channels as the feature map. Each element represents the importance coefficient of the corresponding channel. The weight vector corresponding to a 64-channel feature map contains 64 normalized weight values. The weighting operation is a feature enhancement operation implemented through element-wise multiplication. By expanding the weight vector to a three-dimensional spatial dimension, it is multiplied with the original feature map. For example, a 64×1×1 weight vector is expanded to a 64×7×7 form through a broadcast mechanism and then multiplied element-wise with the original feature map.

[0044] The global average pooling operation compresses features by calculating the mean of spatial dimensions. Its technical effect is to integrate global feature response intensity, avoiding interference from local noise in channel importance assessment. In a specific embodiment, when there are local reflective areas on the copper cable surface, global average pooling balances the contribution ratio of reflective and defective areas by comprehensively calculating the global channel response, preventing excessive weight bias towards high-brightness areas. The combination of the fully connected layer and the nonlinear activation function generates attention weights through two dimensional transformations. Its technical effect is to construct a nonlinear mapping relationship to capture complex feature correlations. For example, when the first fully connected layer compresses a 64-dimensional vector to 16 dimensions, a low-dimensional representation of the potential attention pattern can be extracted. Subsequently, the ReLU activation function suppresses redundant responses. When the second fully connected layer restores the dimensions to 64, a weight distribution that precisely matches the number of channels can be generated. The weighting operation enhances defect-related features through channel-level feature enhancement. Its technical effect is to improve the response intensity of key channels while suppressing irrelevant background information. When detecting a 0.2mm wide scratch, the weight of the edge detection sensitive channel is amplified, and its feature response intensity is significantly improved, while the weight of the background texture channel is suppressed, thereby enhancing the contrast between the defect and the normal area.

[0045] This embodiment provides a machine vision-based method for evaluating the surface quality of high-speed copper cables. It achieves channel-level feature compression by acquiring feature maps and performing global average pooling. Dynamic weight distribution is generated using fully connected layers and nonlinear activation functions. Finally, weighted operations are used to strengthen defect-related features, achieving the technical effects of improving the model's accuracy in identifying subtle defects, reducing computational complexity, and enhancing anti-interference capabilities. Global feature integration ensures that weight generation is unaffected by local noise, the dynamic weight allocation mechanism allows the model to adaptively adjust the focus of feature selection, and channel-level weighted operations improve the feature discrimination ability of the classification layer by suppressing irrelevant information. These techniques work synergistically to maintain a high recognition rate even under complex lighting or motion blur conditions, while meeting the computational efficiency requirements of real-time detection scenarios, providing more reliable feature input data for subsequent defect quantification assessment.

[0046] In one embodiment of the present invention, the defect area ratio is obtained by calculating the ratio of the number of pixels in the defect area to the number of pixels in the entire image. The defect depth estimate is obtained by using a method based on image grayscale changes. By analyzing the grayscale difference between the defect area and the surrounding normal area, and combining the optical characteristic parameters of the copper cable surface material, the following formula is used for calculation:

[0047] Where D is the estimated defect depth, and k is a coefficient related to high-speed copper cables. This represents the grayscale difference between the defective area and the surrounding normal area. This represents the average grayscale value of the normal area. The formula for calculating the percentage of defect area is as follows:

[0048] in, This represents the percentage of the defective area. This represents the number of pixels in the defect area. This represents the total number of pixels in the entire image.

[0049] The number of pixels in the defect area is the total number of valid pixels in the defect area determined by image segmentation technology, obtained through binarization processing or edge detection algorithms. This value includes pixel statistics obtained through thresholding or connected component analysis. The total number of pixels in the entire image is the total number of pixels corresponding to the original resolution output by the acquisition device. For example, the total number of pixels in a 1024×768 resolution image is approximately 786,432. The defect area percentage S is a dimensionless parameter characterizing the extent of defect coverage, with a value ranging from 0 to 1. For example, when the defect area contains 10,000 pixels, the area percentage is approximately 1.27%.

[0050] The defect depth estimate D is a quantitative parameter reflecting the physical depth of the defect, and its calculation depends on the relationship between grayscale difference and optical properties. Grayscale difference It is the absolute difference between the average grayscale value of the defective area and the adjacent normal area. For example, when the grayscale value of the defective area is 80 and that of the normal area is 150... The average gray value of the normal area is 70. It is a local mean calculated using a sliding window or region selection algorithm to eliminate the influence of global illumination variations. The coefficient k related to the copper cable is a calibrated value that combines surface reflectivity, roughness, and light source angle. For example, the k value may be small for polished surfaces, while for textured surfaces, the correspondence between k and actual depth needs to be established experimentally.

[0051] The depth estimation method based on image grayscale changes is implemented through the following steps: First, the mean grayscale values ​​of the defective and normal regions are obtained through sliding window or region segmentation techniques, and the difference between the two is calculated. Secondly, known depth defect samples are calibrated through experiments to establish a mapping relationship between the k value and the actual depth; finally, the D value is calculated by substituting the k values ​​into the formula. This method transforms grayscale differences into actual depth through physical modeling, solving the problem that traditional visual assessment cannot quantify depth, and improves illumination robustness through normalization processing.

[0052] The calculation process for the defect area ratio is achieved through binarization segmentation and pixel statistics: First, a threshold is set for the preprocessed image, marking the defect area as 1 and the background as 0; Next, the image is traversed to count the total number of defective pixels and the total number of pixels, and then the percentage is directly calculated using the formula. This process avoids human visual error and is independent of resolution, ensuring that results are comparable across different devices.

[0053] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. It quantifies defect coverage using pixel ratios, calculates depth values ​​through physical modeling of grayscale differences and material parameters, and ensures robustness to illumination and material adaptability using binarized segmentation and experimental calibration. Furthermore, it optimizes the defect judgment model through comprehensive input of area and depth, thereby improving the scientific rigor of defect assessment, eliminating subjective errors, and expanding application scenarios. This method enables quality reports to include precise quantitative indicators, providing traceable and standardized decision-making basis for production line quality control.

[0054] In one embodiment of the present invention, the formula for calculating the severity assessment value of the defect in the comprehensive judgment model is as follows:

[0055] in, This is a value used to assess the severity of the defect. For defect quantification values, corresponding quantification values ​​are pre-set for different types of defects. This represents the percentage of the defective area. This is an estimate of the defect depth. , and These are the weighting coefficients for the defect quantification value, defect area ratio, and defect depth estimate, respectively, and they satisfy the following conditions: The weighting coefficient , and The settings are determined based on the specifications, application scenarios, and industry standards of high-speed copper cables, and the optimal values ​​are determined through experiments and data analysis.

[0056] Defect Quantification Value This is a numerical representation of the impact of different defect types on the performance of copper cables. Specific values ​​are set through industry standards or expert experience. The defect quantification values ​​include scores corresponding to types such as scratches (e.g., 0.6), pits (e.g., 0.8), and breaks (e.g., 1.0). This value reflects the degree of impact of the defect type on key performance characteristics such as mechanical strength and conductivity. For example, surface oxide spots may be assigned a higher score due to conductivity degradation, while minor scratches will have a lower score.

[0057] Defect area ratio It is the ratio of the area of ​​the defective area to the area of ​​the copper cable surface to be inspected. This parameter is calculated by image processing algorithm and is used to quantify the impact of the defect distribution range on the overall quality. For example, a large area of ​​scratches may be judged as more serious because the coverage area is wide.

[0058] The defect depth estimate D is the depth value in the vertical direction of the defect. This parameter directly reflects the degree of damage the defect causes to the material structure. For example, deep pits may be considered high-risk defects because they reduce mechanical strength.

[0059] Weighting coefficient , and This is a weighted average of the three evaluation parameters, with a sum of 1. The weighting coefficients need to be set in conjunction with the specific specifications and application scenarios of the copper cable. For example, in high-voltage power transmission scenarios, depth has a more significant impact on insulation performance and may improve [the performance of the cable]. Up to 0.4; however, in communication cables, the area ratio of surface defects has a greater impact on signal transmission stability, potentially increasing it. Up to 0.4.

[0060] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. By quantifying multiple dimensions—defect quantification, area ratio, and depth estimation—and combining this with weighted calculations based on dynamically adjusted weight coefficients according to application scenarios, along with a weighting method that uses both experimental data and industry standards for dual verification, the following technical effects can be achieved: First, the linear weighted integration of multi-dimensional parameters overcomes the limitations of traditional methods that rely on a single indicator. For example, by simultaneously considering the hazard level, distribution range, and physical damage degree of defect types, the actual hazard difference between "large-area shallow scratches" and "small-area deep pits" can be distinguished. Second, the dynamic adaptation mechanism of the weight coefficients makes the assessment results more aligned with actual engineering needs. For instance, increasing the area ratio weight in thin-diameter cables enhances sensitivity to surface defects, or emphasizing depth weight in high-voltage scenarios avoids the risk of insulation failure. Finally, the verification through the combination of experimental data and industry standards ensures that the assessment results are consistent with actual failure data and acceptance criteria, thereby improving the scientific rigor and reliability of quality control and providing accurate judgment criteria for real-time monitoring of the production line.

[0061] In one embodiment of the present invention, the specific steps for generating a surface quality assessment report for high-speed copper cables based on defect severity assessment values ​​include: The severity assessment value of a defect is compared with three pre-defined thresholds (first, second, and third) to determine the severity level. The severity assessment value can be a quantitative indicator representing the overall impact of the defect, obtained through weighted calculation of parameters such as defect type, area percentage, and depth. For example, this value can be generated based on the comprehensive judgment result output by an image processing model. The first, second, and third thresholds can be pre-defined grading standards, corresponding to different defect level judgment benchmarks. For example, the first threshold could be 0.3, the second threshold could be 0.6, and the third threshold could be 0.9. The severity level judgment is a classification result determined by comparing the assessment value with the three thresholds step by step. If the assessment value is less than 0.3, it is judged as a minor defect; if it is between 0.3 and 0.6, it is judged as a moderate defect; and if it is greater than or equal to 0.9, it is judged as a severe defect. Through threshold grading technology, the assessment value is sequentially compared with the threshold sequence, and the judgment logic is dynamically adjusted in conjunction with the defect type and quantitative parameters, thereby achieving refined classification of defect levels, avoiding the risk of misjudgment caused by a single threshold, and providing a grading basis for subsequent processing.

[0062] Key data generated throughout the evaluation process is collected, including image acquisition system parameters, defect determination results, defect quantification values, defect area ratios, defect depth estimates, and defect severity assessment values. Image acquisition system parameters can be configuration information of the equipment during operation, including hardware parameters such as camera resolution, light source brightness, and frame rate. Defect determination results are defect type identifiers output by the classification model, including specific defect types such as oxide spots, pits, and scratches. Defect quantification values ​​are numerical values ​​characterizing the intensity of defect features, calculated based on pixel grayscale differences, representing pit depth or scratch length. Defect area ratio is the proportion of the defect area to the overall inspection area, expressed as a percentage. Through key data integration technology, a structured database is used to store equipment parameters, defect types, quantification indicators, and evaluation results in a time- or location-linked manner, ensuring data traceability and providing a complete data chain for quality analysis and process optimization.

[0063] Based on defect severity level determination and key data, corresponding processing suggestions are retrieved from a pre-set defect handling knowledge base. The defect severity level determination, key data, and processing suggestions are then compiled into a surface quality assessment report. The defect handling knowledge base is a rule-based database storing the correspondence between defect levels and processing solutions, including a pre-set rule that triggers a shutdown when a severe pit exceeds 0.2mm in depth. Processing suggestions are measures for handling different defect levels and parameter combinations, including specific operations such as adjusting process parameters, recording and monitoring, or immediate shutdown. Through knowledge base retrieval and report generation technology, using fuzzy matching or precise query methods, the optimal suggestion is retrieved from the knowledge base based on the current defect level and parameter characteristics. Simultaneously, the level determination, parameter records, and suggestion content are integrated into a structured report document, supporting real-time display or database storage. This process combines multi-dimensional information such as defect type, area, and depth, avoiding the limitations of single-indicator decision-making. Furthermore, automated suggestions reduce delays in manual judgment, enabling rapid response in the production process.

[0064] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. It achieves refined defect level determination through multi-level threshold grading, ensures quality traceability through structured data integration, and transforms assessment results into actionable solutions through a knowledge-based suggestion mechanism. This achieves the following technical effects: First, the grading mechanism, combining defect type and quantitative parameters, effectively distinguishes the urgency of defects, avoiding excessive intervention or insufficient response. Second, the complete recording of key data provides a foundation for root cause analysis, such as optimizing the detection environment through correlation analysis between light source brightness and defect type. Third, intelligent suggestions from the knowledge base reduce manual decision-making time, such as automatically triggering a shutdown command when a serious defect is detected. Finally, the linkage between structured reports and the production system forms a closed-loop control, such as automatically adjusting process parameters when a medium-level defect persists, thereby improving production efficiency and product consistency, and promoting the development of industrial inspection towards intelligent automation.

[0065] In one embodiment of the present invention, after generating the surface quality assessment report for high-speed copper cables based on the defect severity assessment value, the method further includes the following step: storing the surface quality assessment report for each assessment in a database. Database storage is a technical means for systematically preserving quality inspection data, implemented through structured data forms. The stored content includes fields such as cable unique identifier, inspection timestamp, defect type classification, defect severity quantification value, and processing suggestions. In a specific embodiment, the database design adopts a relational database management system, ensuring data traceability and analytical compatibility through standardized field definitions.

[0066] Regularly perform statistical analysis on the surface quality assessment reports in the database to generate trend reports on the surface quality of high-speed copper cables. The statistical analysis process uses time series analysis methods to predict the frequency and severity of defects in different time periods through a preset prediction model. The parameters of the prediction model are determined by the maximum likelihood estimation method, and the prediction accuracy of the prediction model is evaluated by the root mean square error.

[0067] Statistical analysis is a technical process based on historical data to identify patterns in quality changes. Predictive models are mathematical tools used to quantify future defect trends, with the ARIMA model used to handle linear trends. Maximum likelihood estimation is a statistical method for determining the optimal solution for model parameters, selecting the autoregressive, differencing, and moving average orders of the ARIMA model by maximizing the probability likelihood function. Root mean square error (RMSE) is a quantitative indicator of prediction accuracy, reflecting the degree of prediction bias through the geometric mean of squared errors. The technical operation is implemented as follows: First, structured time-series data is extracted from the database, and moving averages or seasonal decomposition methods are used to identify trends in defect frequency and severity. Second, a predictive model is selected based on data characteristics, and maximum likelihood estimation is used to determine parameter combinations. Third, historical data is divided into training and validation sets, and iterative optimization is performed to ensure the model's RMSE on the test set is below a preset threshold. Finally, the optimized model is deployed to generate prediction results and warning thresholds for future periods. This process enables proactive identification of defect trends, such as adjusting equipment maintenance plans by detecting a rise in defect rates every Friday afternoon, or predicting defect rate fluctuations caused by material expansion during high-temperature seasons.

[0068] The formula for calculating the root mean square error is:

[0069] in, Let n be the root mean square error, and n be the number of samples. This is the actual value. This is the predicted value. The formula transforms discrete prediction deviations into a quantifiable accuracy index by calculating the root mean square error of the squared error. In a specific embodiment, if a prediction model's RMSE for predicting the number of defects in the past 30 days is 0.05, it indicates that the average relative deviation between its predicted and actual values ​​is controlled within 5%, meeting the reliability requirements of industrial scenarios. The introduction of this index provides a quantifiable evaluation standard for the model optimization process; for example, the optimal prediction scheme can be selected by comparing the RMSE values ​​of different models.

[0070] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. By structurally storing assessment reports in a database to form data assets, combining time series analysis to identify defect evolution patterns, and utilizing parameter-optimized predictive models to achieve forward-looking projections of quality trends, the method employs root mean square error (RMSE) to quantify model accuracy and drive iterative optimization. This achieves the technical effect of constructing a closed-loop quality monitoring system. This system accumulates knowledge through historical data, uses statistical analysis to identify systemic quality risks, and leverages predictive models to transform passive detection into proactive intervention. Ultimately, it achieves dynamic optimization of the production process and defect prevention, significantly improving product qualification rates and production continuity.

[0071] In one embodiment, the specific determination logic for the severity level of the defect is as follows: When the defect severity assessment value is 0, it is determined that there is no defect. The defect severity assessment value can be a quantitative indicator obtained by feature extraction and comprehensive analysis of copper cable surface images through a machine vision system. It can be obtained by weighting parameters such as defect type, area, and depth using a convolutional neural network model. For example, the assessment value can include a comprehensive score based on parameters such as pixel anomaly degree and texture feature deviation.

[0072] When the defect severity assessment value is greater than 0 and less than or equal to the first threshold, it is determined that there is a minor defect. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to strengthen monitoring in subsequent production processes. The first threshold is a critical value determined based on statistical analysis of historical defect cases, set through cluster analysis or domain expert experience. Minor defects are anomalies that have a minimal impact on product performance, such as tiny oxide spots or minor scratches. Enhanced monitoring can be achieved by adjusting image acquisition frequency or increasing the proportion of manual spot checks.

[0073] When the defect severity assessment value is greater than the first threshold and less than or equal to the second threshold, it is determined that there is a general defect. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to adjust the production process parameters. The second threshold is an intermediate critical value based on the potential impact of defects on product functionality, determined through regression analysis or sensitivity testing of process parameters. Adjusting production process parameters includes specific operations such as reducing production line speed, optimizing light source angle, or adjusting coating thickness.

[0074] When the defect severity assessment value is greater than the second threshold and less than or equal to the third threshold, a serious defect is determined to exist. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to stop production immediately and inspect and repair the relevant equipment. The third threshold is a critical value based on whether a defect could lead to significant quality risks or safety hazards. Immediately stopping production can trigger automated control commands, such as directly shutting down the production line via a programmable logic controller.

[0075] When the defect severity assessment value is greater than the third threshold, it is determined that there is an extremely serious defect. At this time, the handling suggestions in the defect handling knowledge base include: in addition to stopping production and inspecting the equipment, a comprehensive review of the copper cables that have been produced is also required. Critically severe defects are extreme cases where defects could lead to batch-wide quality incidents or customer complaints, such as pits deeper than 0.15 mm or large-area plating peeling. A comprehensive review can include batch-by-batch screening of manufactured products using X-ray inspection or tensile testing.

[0076] The defect handling knowledge base was established through the analysis and summarization of a large number of historical defect cases.

[0077] The historical defect cases include structured storage of past inspection data, manual re-inspection records, and customer feedback information. For example, case data includes multi-dimensional information such as defect type, occurrence time, handling measures, and effects. The knowledge base construction process may include data collection, pattern recognition, rule generation, and verification optimization. Data collection is achieved through a database interface, pattern recognition is completed through cluster analysis and mining, rule generation is based on expert experience or machine learning model output, and verification optimization is conducted through A / B testing or expert review.

[0078] The specific technical operations are implemented in the following ways: After the defect severity assessment value is input into the judgment module, the system uses an interval matching algorithm to determine the level range to which the current assessment value belongs. The processing suggestion retrieval then extracts pre-defined structured rules from the knowledge base based on the matching results. For example, when a defect is judged to be severe, the system invokes the "Stop for Inspection" rule and associates it with specific suggestions for the current defect type. The report generation module integrates the level judgment, suggestions, and key data into a visual report and triggers alarms or control commands through a message queue or API interface. The dynamic update mechanism imports new case data periodically and retrains the rule base. For example, when it is found that adjusting the light source brightness can reduce the false detection rate, the system automatically updates the relevant rule parameters.

[0079] This embodiment provides a machine vision-based method for assessing the surface quality of high-speed copper cables. By dividing the defect severity assessment value into five quantification intervals and establishing corresponding processing suggestion rules, and combining this with a defect processing knowledge base constructed from historical cases and expert experience, an interval matching algorithm is used to automatically associate defect level determination with processing suggestions. A dynamic update mechanism enhances the adaptability of the knowledge base, achieving the following technical effects: 1. Avoid over- or under-response through a refined classification strategy. For example, after distinguishing between minor and general defects, the monitoring frequency can be adjusted accordingly instead of directly shutting down the machine, thereby reducing production interruption losses. 2. A structured rule base based on historical cases and expert experience reduces the risk of human error. For example, the knowledge base can be used to correct operators' misjudgments of spiral patterns or to optimize processing suggestions for new coating materials. 3. The collaborative optimization of the knowledge base and upstream steps (such as defect classification models and prediction models) improves the system's intelligence level. For example, it can link the PLC to automatically adjust the production line speed or trigger secondary verification rules. 4. The comprehensive review and equipment inspection recommendations' tiered triggering mechanism effectively controls batch risks, such as tracing and isolating batches with deep defects caused by mold wear; 5. The dynamic update mechanism enables the system to adapt to process improvements or the introduction of new materials. For example, the optimal process parameter range can be iteratively updated by analyzing the effects of "adjusting parameters" suggestions.

[0080] In summary, this method achieves an upgrade from isolated detection to intelligent quality control, significantly improving the automation level of the production line and product consistency.

[0081] See Figure 2 A machine vision-based high-speed copper cable surface quality assessment system includes: The image acquisition module is used to acquire images of the surface of high-speed copper cables. It consists of an image acquisition unit composed of several high-speed cameras and a uniform ring light source. Multiple high-speed CCD cameras are arranged in a ring array, with dynamically adjustable frame rates, supporting an ultra-high-speed acquisition mode of up to 2000fps. The camera installation angle has been optimized through optical simulation, with the angle between the optical axes of adjacent cameras being 45°±5°, ensuring an overlap rate of ≥20% in the shooting angles, achieving 360° coverage of the copper cable surface without blind spots. The ring camera array and overlapping angle design completely eliminate the blind spot problem of traditional single-sided camera detection, improving the detection coverage of the copper cable surface and avoiding the risk of missed detections.

[0082] The image preprocessing module, electrically connected to the image acquisition module, is used to receive and preprocess the acquired image data. It is equipped with a grayscale conversion unit, a filtering and denoising unit, an edge detection unit, and a morphological processing unit. The defect classification module is equipped with a defect classification model. It is used to receive the preprocessed image data, classify and judge defects, and output the defect judgment result and defect quantification value. The defect assessment module is equipped with a comprehensive judgment model, which calculates the severity assessment value of the defect based on the quantitative value of the defect judgment result, the defect area ratio, and the defect depth estimate. This allows for the determination of the severity of the surface defects of the copper cable and the assessment of the surface quality of the copper cable, generating an assessment report. The defect assessment module has built-in weighting coefficients set according to the specifications of the copper cable, application scenarios, and industry standards, as well as thresholds determined according to the enterprise's quality control requirements and production experience. The data storage module is equipped with an analysis unit to store evaluation reports and perform periodic statistical analysis to generate copper cable surface quality trend reports.

[0083] This invention achieves high-speed, clear acquisition and full surface coverage of copper cable surface images using a high-speed camera and a ring light source. After adaptive preprocessing to improve image quality, an improved deep learning model combined with an attention mechanism is used to accurately identify subtle defects. Then, a multi-parameter comprehensive judgment model and dynamic threshold calibration are used to quantitatively assess the severity of defects. Finally, distributed data storage and an integrated learning model are used for trend analysis and prediction, thereby constructing an automated and intelligent surface quality assessment system. This significantly improves defect identification accuracy and detection efficiency, provides scientific and quantitative quality assessment reports and decision support for the production process, and effectively solves the problems of insufficient high-speed detection accuracy and lack of quantitative assessment in existing technologies.

[0084] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A machine vision-based method for evaluating the surface quality of high-speed copper cables, characterized in that, Includes the following steps: An image acquisition system is built by using a camera that surrounds a high-speed copper cable and a uniform ring light source to acquire image data of the high-speed copper cable. The image data is preprocessed, which includes image grayscale conversion, filtering and noise reduction, edge detection and morphological processing in sequence. Establish a surface defect classification model based on deep learning, input preprocessed image data, and output the defect judgment result and defect quantification value of the image data; Based on the preprocessed image data and defect judgment results, the defect area ratio and defect depth estimate are calculated. The defect quantification value, defect area ratio and defect depth estimate are input into the comprehensive judgment model to calculate the defect severity assessment value. Based on the defect severity assessment value, a surface quality assessment report of high-speed copper cable is generated.

2. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 1, characterized in that, The image grayscale conversion specifically includes: converting the color image of the image data into a grayscale image; the filtering and denoising specifically includes: applying a Gaussian filtering algorithm to denoise the grayscale image; the edge detection specifically includes: using an edge detection algorithm to extract edges from the denoised image to obtain the image edge information of the high-speed copper cable; the morphological processing specifically includes: using dilation and erosion operations to optimize the edge-extracted image; wherein, the size of the Gaussian kernel in the Gaussian filtering algorithm is adaptively adjusted according to the surface texture complexity of the high-speed copper cable.

3. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 1, characterized in that, The surface defect classification model adopts an improved convolutional neural network structure with an attention mechanism module. When establishing the surface defect classification model, the model is trained using high-speed copper cable image data containing different types of defects. The learning rate during model training adopts a dynamic adjustment strategy, which specifically involves setting an initial learning rate and gradually decreasing it as the number of training rounds increases.

4. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 3, characterized in that, When establishing a surface defect classification model, the specific working process of the attention mechanism module is as follows: Obtain a feature map of the input image data, the feature map having three dimensions: number of channels, height, and width; Global average pooling is performed on the feature map to compress it in the height and width dimensions, resulting in a one-dimensional vector that retains only the channel dimension information. The one-dimensional vector is sequentially input into two fully connected layers, and a non-linear activation function is used between the two fully connected layers to process it, thereby obtaining the attention weight vector, the dimension of which is consistent with the number of channels; The attention weight vector is weighted with the input feature map to enhance the feature information related to defects in the feature map, and the weighted feature map is output.

5. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 1, characterized in that, The defect area ratio is obtained by calculating the ratio of the number of pixels in the defect area to the number of pixels in the entire image. The defect depth estimate is obtained by using a method based on image grayscale changes. By analyzing the grayscale difference between the defect area and the surrounding normal area, and combining the optical property parameters of the copper cable surface material, the estimate is calculated using the following formula: Where D is the estimated defect depth, and k is a coefficient related to high-speed copper cables. This represents the grayscale difference between the defective area and the surrounding normal area. This represents the average grayscale value of the normal area. The formula for calculating the percentage of defect area is as follows: in, This represents the percentage of the defective area. This represents the number of pixels in the defect area. This represents the total number of pixels in the entire image.

6. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 1, characterized in that, The formula for calculating the severity assessment value of defects in the comprehensive judgment model is as follows: in, This is a value used to assess the severity of the defect. For defect quantification values, corresponding quantification values ​​are pre-set for different types of defects. This represents the percentage of the defective area. This is an estimate of the defect depth. , and These are the weighting coefficients for the defect quantification value, defect area ratio, and defect depth estimate, respectively, and they satisfy the following conditions: The weighting coefficient , and The settings are determined based on the specifications, application scenarios, and industry standards of high-speed copper cables, and the optimal values ​​are determined through experiments and data analysis.

7. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 1, characterized in that, The specific steps for generating a surface quality assessment report for high-speed copper cables based on defect severity assessment values ​​include: The severity assessment value of the defect is compared with a pre-set first threshold, second threshold, and third threshold to obtain the severity level of the defect. Collect key data generated throughout the evaluation process, including the acquisition parameters of the image acquisition system, defect determination results, defect quantification values, defect area ratio, defect depth estimation values, and defect severity assessment values. Based on the severity level of defects and key data, corresponding processing suggestions are obtained from a pre-set defect processing knowledge base. The severity level of defects, key data and processing suggestions are then summarized into a surface quality assessment report.

8. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 1, characterized in that, After generating the surface quality assessment report for high-speed copper cables based on the defect severity assessment value, the following steps are also included: Store the surface quality assessment report for each evaluation in the database; Regularly perform statistical analysis on the surface quality assessment reports in the database to generate trend reports on the surface quality of high-speed copper cables. The statistical analysis process uses time series analysis methods to predict the frequency and severity of defects in different time periods through a preset prediction model. The parameters of the prediction model are determined by the maximum likelihood estimation method, and the prediction accuracy of the prediction model is evaluated by the root mean square error. The formula for calculating the root mean square error is: in, Let n be the root mean square error, and n be the number of samples. This is the actual value. These are predicted values.

9. The machine vision-based surface quality assessment method for high-speed copper cables according to claim 7, characterized in that, The specific determination logic for the severity level of the defect is as follows: When the defect severity assessment value is 0, it is determined that there is no defect. When the defect severity assessment value is greater than 0 and less than or equal to the first threshold, it is determined that there is a minor defect. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to strengthen monitoring in subsequent production processes. When the defect severity assessment value is greater than the first threshold and less than or equal to the second threshold, it is determined that there is a general defect. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to adjust the production process parameters. When the defect severity assessment value is greater than the second threshold and less than or equal to the third threshold, a serious defect is determined to exist. At this time, the handling suggestions in the defect handling knowledge base include: it is recommended to stop production immediately and inspect and repair the relevant equipment. When the defect severity assessment value is greater than the third threshold, it is determined that there is an extremely serious defect. At this time, the handling suggestions in the defect handling knowledge base include: in addition to stopping production and inspecting the equipment, a comprehensive review of the copper cables that have been produced is also required. The defect handling knowledge base was established through the analysis and summarization of a large number of historical defect cases.

10. A machine vision-based high-speed copper cable surface quality assessment system, used to implement the method described in any one of claims 1-9, characterized in that, include: The image acquisition module is used to acquire images of the surface of high-speed copper cables. It includes an image acquisition unit composed of several high-speed cameras and a uniform ring light source. The image preprocessing module is electrically connected to the image acquisition module. It is used to receive the acquired image data and perform preprocessing. It is equipped with a grayscale conversion unit, a filtering and noise reduction unit, an edge detection unit, and a morphological processing unit. The defect classification module is equipped with a defect classification model, which is used to receive preprocessed image data, classify and judge defects, and output defect judgment results and defect quantification values. The defect assessment module is equipped with a comprehensive judgment model, which calculates the severity assessment value of the defect based on the quantitative value of the defect judgment result, the defect area ratio, and the defect depth estimate. This allows for the determination of the severity of the surface defects of the copper cable and the assessment of the surface quality of the copper cable, generating an assessment report. The defect assessment module has built-in weighting coefficients set according to the specifications of the copper cable, application scenarios, and industry standards, as well as thresholds determined according to the enterprise's quality control requirements and production experience. The data storage module is equipped with an analysis unit to store evaluation reports and perform periodic statistical analysis to generate copper cable surface quality trend reports.

Images