A cutting quality assessment method and system for automobile caliper production line

Abstract

The present invention provides a cutting quality assessment method and system for an automobile caliper production line, which relates to the field of automobile manufacturing technology, including collecting image data of the cutting surface of the automobile caliper at different use stages; extracting the first texture feature and the second texture feature of the cutting surface of the automobile caliper; constructing and training an autoencoder model; obtaining a comprehensive feature representation through feature fusion processing; processing the comprehensive feature representation through feature selection based on mutual information to obtain the feature that contributes most to the judgment of the cutting surface state, and obtaining the classification result of the cutting surface state through long short-term memory convolutional neural network model training and optimization processing. The present invention realizes rapid and accurate assessment of the cutting surface of the automobile caliper, real-time monitoring and assessment of the cutting surface condition, provides a scientific basis for the production and quality control of the automobile caliper, and ensures the safety and reliability of the automobile.

Description

Technical Field

[0001] The present invention relates to the technical field of automobile manufacturing, and in particular to a cutting quality assessment method and system for an automobile caliper production line. Background Art

[0002] In terms of the cutting quality assessment of the cutting surface in the automobile caliper production line, the current state of technology has some shortcomings, as follows: the application of traditional manual inspection methods in the automotive industry has increasingly highlighted the problems of low efficiency and insufficient inspection accuracy. With the development of a new generation of information technology, intelligent manufacturing has become a trend, and the demand for automated and intelligent visual inspection systems is growing; the existing technology still needs to be improved in terms of the degree of automation, and many production links still rely on manual operation, which not only affects production efficiency, but also makes it difficult to ensure the consistency of processing quality; traditional surface quality inspection technologies, such as eddy current inspection, AC electromagnetic field inspection, magnetic leakage inspection and laser ultrasonic inspection, although they have certain applications in specific fields, they are limited in applicable occasions, defect type identification and quantitative description, and cannot meet the current surface defect detection needs of industrial products.

[0003] Therefore, traditional surface defect detection methods such as eddy current detection and AC electromagnetic field detection are only applicable to specific materials and specific types of defects, and have high requirements for surface finish, making it difficult to adapt to diverse industrial detection needs. Existing automated detection systems often lack deep integration with production processes, leading to data islands and making it difficult to achieve comprehensive quality control and traceability. In addition, with the diversification of product types, existing detection systems are difficult to quickly adapt to new detection needs and lack flexibility and scalability. In summary, the existing technology for cutting quality assessment of automobile caliper production lines has problems such as insufficient automation and backward surface quality detection technology. Summary of the invention

[0004] The purpose of the present invention is to provide a cutting quality assessment method and system for an automobile caliper production line to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0005] In a first aspect, the present application provides a method for evaluating the cutting quality of an automobile caliper production line, comprising:

[0006] Collect image data of the cutting surface of the automobile caliper at different stages of use. The image data at each stage includes image information of the glossiness, color uniformity, and surface scratches of the cutting surface;

[0007] Among them, obtaining image data: This step ensures that baseline data on the glossiness, color uniformity and surface integrity of the cutting surface can be obtained. After the caliper undergoes routine maintenance, it is photographed again using the same image acquisition device to record the impact of maintenance on the state of the cutting surface. This step helps to evaluate the maintenance effect of the maintenance process on the quality of the cutting surface. For calipers that have been used for a long time, the surface and aging of the cutting surface are captured by an image acquisition device. Through the above image acquisition steps, detailed image data reflecting the state of the cutting surface of the automotive caliper at different stages of use can be obtained. These data include not only the glossiness and color uniformity of the cutting surface, but also surface scratches or image information of the cutting surface.

[0008] The image data is processed using a Gabor filter to extract the first texture feature and the second texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution characteristics of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, and the second texture feature is a Gabor filter of different scales, each scale is designed to capture the texture details of the cutting surface within a frequency range;

[0009] Based on the first texture feature and the second texture feature, an autoencoder model is constructed and trained, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing reconstruction errors, wherein the encoder compresses high-dimensional image data into low-dimensional feature representations, and the decoder reconstructs the original image from the low-dimensional features, thereby providing deep learning extracted feature representations for cutting surface evaluation of the cutting surface of the automobile caliper;

[0010] According to the first texture feature, the second texture feature and the feature representation extracted by deep learning, a comprehensive feature representation is obtained through feature fusion processing, wherein the feature fusion processing includes weighted average, PCA dimension reduction and Dense layer fusion method of deep learning;

[0011] The comprehensive feature representation is processed by feature selection based on mutual information to obtain the feature that contributes most to the judgment of the cutting surface state, wherein the feature selection process includes calculating the mutual information between the feature and the cutting surface state, and selecting the feature with the largest mutual information as an input item;

[0012] Based on the most contributing features, the classification results of the cutting surface status are obtained through long short-term memory convolutional neural network model training and optimization processing. The model training and optimization include the use of cross-validation, adjustment of network structure and selection of optimization algorithm. The classification results of the cutting surface status are used to evaluate the condition of the cutting surface of automobile calipers.

[0013] Preferably, the image data is processed using a Gabor filter to extract the first texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution characteristics of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, including:

[0014] Design Gabor filters for the cutting surface of the car caliper in different directions, including 0°, 45°, 90° and 135° directions;

[0015] The designed Gabor filter is applied to the image data of the car caliper surface, where a convolution operation is performed for the filter in each direction;

[0016] Based on the convolution operation results, the features of roughness, granularity and crack depth are extracted from the filtered image.

[0017] Preferably, the autoencoder model is constructed and trained, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing the reconstruction error, wherein the encoder compresses the high-dimensional image data into low-dimensional feature representations, and the decoder reconstructs the original image from these low-dimensional features, thereby providing deep learning extracted feature representations for the evaluation of the cutting surface of the automobile caliper, including:

[0018] Design an autoencoder model that includes an encoder and a decoder, where the encoder consists of multiple convolutional layers and pooling layers to extract image features, and the decoder consists of the transpose operation of the convolutional layer to reconstruct the image;

[0019] After being processed by the autoencoder model, the high-dimensional image data is compressed into a low-dimensional feature vector, where the low-dimensional feature vector contains the main information of the image;

[0020] According to the low-dimensional feature vector, after being processed by a decoder, a reconstructed image is obtained, wherein the reconstructed image is an approximate representation of the original image data, and a high-level feature representation of the data is automatically learned by optimizing the reconstruction error, wherein the optimized reconstruction error includes the difference between the original image and the reconstructed image;

[0021] According to the training results of the autoencoder model, after feature selection and dimensionality reduction processing, a low-dimensional feature representation that can represent the key information of the image is obtained, and finally a feature representation extracted by deep learning for subsequent cutting surface evaluation is obtained.

[0022] Preferably, the feature representation extracted according to the first texture feature, the second texture feature and deep learning is subjected to feature fusion processing to obtain a comprehensive feature representation;

[0023] Fusing the low-dimensional features obtained by the autoencoder with the first physical characteristic;

[0024] The final feature representation is subjected to feature normalization, feature selection and dimensionality reduction to obtain an optimized and comprehensive feature representation.

[0025] Preferably, the comprehensive feature representation is processed by feature selection based on mutual information to obtain the feature that contributes most to the judgment of the surface state of the cutting surface, wherein the feature selection process includes calculating the mutual information between the feature and the surface state of the cutting surface, and selecting the feature with the largest mutual information as an input item, including:

[0026] Let X be the set of features extracted from the comprehensive feature representation, let Y be the cutting surface state label of the automobile caliper, determine the probability of each feature and the cutting surface state appearing at the same time, calculate the probability of each feature appearing alone and the probability of each cutting surface state appearing alone, and calculate the mutual information between the feature and the cutting surface state;

[0027] The mutual information of each feature is evaluated, and the features are sorted. The feature with the largest mutual information is selected as the feature that contributes most to the judgment of the surface state of the cutting surface.

[0028] In a second aspect, the present application also provides a cutting quality assessment system for an automobile caliper production line, comprising:

[0029] Acquisition module: used to collect image data of the cutting surface of the automobile caliper at different use stages. The image data at each stage includes the image information of the glossiness, color uniformity, and surface scratches of the cutting surface;

[0030] Extraction module: used to process the image data using Gabor filters, extract the first texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution feature of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, and extract the second texture feature of the cutting surface of the automobile caliper, wherein the second texture feature is Gabor filters of different scales, and each scale is designed to capture the texture details of the cutting surface within a frequency range;

[0031] A construction module: used for constructing and training an autoencoder model based on the first texture feature and the second texture feature, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing the reconstruction error, wherein the encoder compresses the high-dimensional image data into a low-dimensional feature representation, and the decoder reconstructs the original image from the low-dimensional features, thereby providing a feature representation extracted by deep learning for the evaluation of the cutting surface of the automobile caliper;

[0032] The first processing module is used to obtain a comprehensive feature representation through feature fusion processing based on the first texture feature, the second texture feature and the feature representation extracted by deep learning, wherein the feature fusion processing includes weighted averaging, PCA dimensionality reduction and Dense layer fusion method of deep learning;

[0033] The second processing module is used to process the comprehensive feature representation through feature selection based on mutual information to obtain the feature that contributes most to the judgment of the surface state of the cutting surface, wherein the feature selection process includes calculating the mutual information between the feature and the surface state of the cutting surface, and selecting the feature with the largest mutual information as an input item;

[0034] Optimization module: It is used to obtain the classification results of the cutting surface state based on the most contributing features after long short-term memory convolutional neural network model training and optimization processing. The model training and optimization include the use of cross-validation, adjustment of network structure and selection of optimization algorithm. The classification results of the cutting surface state are used to evaluate the cutting surface condition of the cutting surface of the automobile caliper.

[0035] In a third aspect, the present application also provides a cutting quality assessment device for an automobile caliper production line, comprising:

[0036] Memory for storing computer programs;

[0037] A processor is used to implement the steps of the cutting quality assessment method for the automobile caliper production line when executing the computer program.

[0038] In a fourth aspect, the present application further provides a readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned cutting process quality assessment method based on the automobile caliper production line are implemented.

[0039] The beneficial effects of the present invention are:

[0040] The present invention uses a long short-term memory convolutional neural network model, combined with cross-validation, network structure adjustment and optimization algorithm selection, so that the present invention can more accurately evaluate the cutting surface state of the cutting surface of the automobile caliper, capture the subtle features of the cutting surface surface, including roughness, granularity and crack depth, thereby improving the accuracy of the evaluation; the k-fold cross-validation method is adopted to ensure that the model has good performance on different data subsets, which helps to improve the generalization ability of the model and ensure stability and reliability in practical applications.

[0041] The present invention reduces the time and resources required for model training and improves computational efficiency by adjusting the network structure and selecting a suitable optimization algorithm, evaluates the cutting surface of automobile calipers more quickly, and accelerates the detection speed of the production line; multi-source features are effectively fused through weighted averaging, PCA dimensionality reduction, and deep learning Dense layer fusion method, which not only retains the most contributing features but also removes redundant information, making the feature representation more comprehensive and compact.

[0042] The model of the present invention can be applied to the automobile caliper production line in real time to quickly evaluate the cutting surface status, provide decision support for timely maintenance and replacement, thereby improving production efficiency and product quality. Through the automated and intelligent cutting surface evaluation method, the present invention reduces the need for manual inspection, reduces labor costs and the risk of misjudgment due to human factors; accurately evaluating the cutting surface surface status of the automobile caliper cutting surface is crucial to ensuring the automobile braking performance, and helps to improve the safety of the automobile and extend its service life.

[0043] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by implementing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0045] Figure 1 It is a schematic flow chart of a cutting quality assessment method for an automobile caliper production line according to an embodiment of the present invention;

[0046] Figure 2 It is a schematic diagram of the structure of the cutting quality evaluation system of the automobile caliper production line described in an embodiment of the present invention;

[0047] Figure 3 It is a schematic diagram of the structure of the cutting quality assessment equipment for the automobile caliper production line described in an embodiment of the present invention.

[0048] In the figure: 701, acquisition module; 702, extraction module; 703, construction module; 704, first processing module; 705, second processing module; 706, optimization module; 800, cutting quality assessment equipment for automobile caliper production line; 801, processor; 802, memory; 803, multimedia component; 804, I / O interface; 805, communication component. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0050] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance. Embodiment 1:

[0051] This embodiment provides a method for evaluating the cutting quality of an automobile caliper production line.

[0052] See also Figure 1 , the figure shows that the method includes step S100, step S200, step S300, step S400, step S500 and step S600.

[0053] S100, collecting image data of the cutting surface of the automobile caliper at different use stages, wherein the image data at each stage includes image information of glossiness, color uniformity, and surface scratches of the cutting surface.

[0054] It can be understood that in this step, the bevel gear images with different cutting surface degrees at different usage stages can be used for target detection model training related to machining quality control and automated inspection, which is crucial for evaluating the generalization performance of the model for calipers with improved geometric features, especially in the recognition and feature description performance of tiny defects. The image data will show the original appearance of the cutting surface, including gloss, color uniformity, etc.; by collecting image data at different stages, the surface condition of the cutting surface can be fully understood, providing accurate data support for subsequent cutting surface surface evaluation; and by analyzing these image data, the performance changes of the cutting surface at different usage stages can be understood, thereby evaluating its wear resistance and durability.

[0055] S200. Process the image data using a Gabor filter to extract a first texture feature and a second texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is a distribution feature of the roughness, granularity, and crack depth of the cutting surface in directions of 0°, 45°, 90°, and 135°, and the second texture feature is a Gabor filter of different scales, each scale being designed to capture texture details of the cutting surface within a frequency range.

[0056] It can be understood that in this step, high-resolution surface image data of the cutting surface is obtained through the image acquisition device. According to the acquired image data, preprocessing operations are performed, including grayscale, contrast enhancement, etc., to highlight the texture features of the cutting surface, and a preprocessed image for further analysis is obtained, and texture features are extracted using Gabor filters. First, a series of Gabor kernels of different directions (0°, 45°, 90° and 135°) and different scales are constructed, each kernel is designed for a specific frequency and direction to match the texture features that may appear on the cutting surface, wherein a series of Gabor filter kernels are designed, each kernel corresponding to a specific direction i and scale s, The Gabor filter outputs of different directions and scales are fused to obtain a comprehensive texture feature vector. The principal component analysis (PCA) is used to reduce the dimension of the feature and retain the main components. The formula is as follows: PCA(X)=XV Where, X is the feature matrix, V is the eigenvector obtained through PCA analysis;

[0057] Based on the image processed by the Gabor filter, the first texture features are calculated, including the distribution characteristics of roughness, granularity and crack depth in the specified direction. This step involves thresholding the filtered image to facilitate the analysis and quantification of the physical properties of the cutting surface. Based on the different scale outputs of the Gabor filter, the second texture features are extracted, that is, the texture details of the cutting surface in different frequency ranges, which includes power spectrum analysis of the filtered image to identify the frequency distribution characteristics of the cutting surface texture at different scales.

[0058] Next, based on the extracted texture features, feature fusion is performed to integrate the directional and scale-frequency features into a comprehensive feature vector. This step may involve feature dimension reduction techniques, such as principal component analysis (PCA), to reduce the dimension of the data and extract the most representative features. After the feature selection process, a mutual information-based method is used to evaluate the correlation between each feature and the surface state of the cutting surface, and the feature with the largest mutual information is selected as the final evaluation feature of the cutting surface.

[0059] It should be noted that the distribution characteristics of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135° are extracted because these directions cover all the main directions in which the caliper cutting surface may be subjected to the cutting surface. During the actual use of the car, the caliper will be subjected to forces and friction from multiple directions. In addition, 0° usually represents the horizontal direction, 45° and 135° represent the diagonal directions, and 90° represents the vertical direction, which helps to analyze the physical properties of the cutting surface in different directions.

[0060] Wherein S200 includes S201, S202 and S203, wherein:

[0061] S201, designing Gabor filters on the cutting surface of the automobile caliper in different directions, wherein the different directions include 0°, 45°, 90° and 135° directions, and the calculation formula is as follows: Where G is the filter, x 'and y ′ is the coordinate after rotation, l is the wavelength, i is the direction of the filter, ψ is the phase shift, s is the standard deviation of the Gaussian kernel, c is the spatial aspect ratio;

[0062] S202, applying the designed Gabor filter to the image data of the automobile caliper surface, wherein for the filter in each direction, a convolution operation is performed, and the calculation formula is as follows: In the formula, I ′(x , y ) is the filtered image, I ( x , y ) is the original image in spatial coordinates ( x , y ), G ( x , y , l , i , ψ , s , c ) is a complex function, which is the functional expression of the Gabor filter;

[0063] S203: Based on the convolution operation result, extract the features of roughness, granularity and crack depth from the filtered image.

[0064] It should be noted that roughness: calculates the local variance of the filtered image to assess surface irregularities: Where I′ˉ is the average intensity of the filtered image, and N is the total number of pixels;

[0065] Granularity: It is evaluated by analyzing the contrast and uniformity of the gray-level co-occurrence matrix (GLCM) of the filtered image: Granularity = GLCM(I′,θ) where GLCM calculates the co-occurrence frequency of pixel pairs in the direction θ.

[0066] Crack Depth: Estimate the depth and width of cracks by detecting high-frequency components in the filtered image: In the formula, Δ x and Δ yis a differential operator in the x and y directions. S300, based on the first texture feature and the second texture feature, construct and train an autoencoder model, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing the reconstruction error, wherein the encoder compresses the high-dimensional image data into a low-dimensional feature representation, and the decoder reconstructs the original image from these low-dimensional features, thereby providing a deep learning extracted feature representation for the evaluation of the cutting surface of the automobile caliper. Based on the extracted features, it is determined whether the cutting surface has defects by setting a threshold or using a machine learning model. For example, the Gabor filter in the OpenCV library can be used to implement texture extraction and defect analysis, and the abnormal area on the cutting surface surface can be identified by analyzing the filtered image. For the detected defects, the defect type is further classified and identified, such as scratches, dents, rust, etc. This can be achieved by training a deep learning model that can identify different defect types based on feature vectors. The defect detection results, including the location, type and severity of the defects, are output to provide a basis for subsequent quality control and maintenance, which can be achieved through an automated system that can feedback defect information in real time to improve production efficiency and product quality.

[0067] It can be understood that step S300 includes S301, S302, S303 and S304, wherein:

[0068] S301. Design an autoencoder model including an encoder and a decoder, wherein the encoder is composed of multiple convolutional layers and pooling layers for extracting image features, and the decoder is composed of a transposition operation of the convolutional layer for reconstructing the image;

[0069] S302, compressing the high-dimensional image data into a low-dimensional feature vector after being processed by the autoencoder model, wherein the low-dimensional feature vector contains the main information of the image;

[0070] It should be noted that, according to the high-dimensional image data, after multiple convolutional layers and pooling layers, a low-dimensional feature representation is obtained, where the convolutional layer is used to extract the local features of the image, and the pooling layer is used to reduce the spatial size of the features and increase the abstract level of the features. Then, according to the low-dimensional features output by the encoder, after a series of transposition operations of the convolutional layers, the image is reconstructed, where the transposed convolutional layer (also called deconvolution or upsampling convolution) is used to increase the spatial size of the feature map and reconstruct the details of the image. Its calculation formula is as follows: In the formula, is the output feature of the previous layer, represents the transposed convolution operation of layer 1, Dlayer represents the output of the transposed convolutional layer. In step S302, a compressed low-dimensional feature vector is obtained based on the feature map processed by the encoder. This step achieves dimensionality reduction of the image data, extracts the main features of the image, and removes some unimportant information. Based on the low-dimensional feature vector, an image close to the original image is reconstructed after processing by the decoder. This step is the core of the autoencoder, which attempts to reconstruct the original data through the learned low-dimensional features, thereby verifying the validity of the features and the generalization ability of the model.

[0071] In this embodiment, the encoder part of the autoencoder is used to achieve effective compression of image data, extract low-dimensional feature vectors containing the main information of the image, reduce the dimension of the data, and retain the key information of the original data as much as possible, which is convenient for subsequent analysis and processing. Through the decoder part of the autoencoder, effective reconstruction of the compressed features is achieved. Although the reconstructed image may be different from the original image, it can approximate the main features and structure of the original image well. In practical applications, the autoencoder model can be applied to multiple fields such as image compression, feature extraction, and image denoising.

[0072] S303, obtaining a reconstructed image through a decoder processing according to the low-dimensional feature vector, wherein the reconstructed image is an approximate representation of the original image data, and automatically learning a high-level feature representation of the data by optimizing a reconstruction error, wherein the optimized reconstruction error includes a difference between the original image and the reconstructed image;

[0073] S304. According to the training results of the autoencoder model, after feature selection and dimensionality reduction processing, a low-dimensional feature representation that can represent the key information of the image is obtained, and finally a feature representation extracted by deep learning for subsequent cutting surface evaluation is obtained.

[0074] It should be noted that the low-dimensional feature vector is the output of the encoding stage of the autoencoder, representing the compressed representation of the input image data. The decoder consists of a series of transposed convolutional layers that gradually upsample ZZ and finally reconstruct the image. The decoding process can be expressed as: in, D last represents the last layer of the decoder, which usually uses a Sigmoid activation function to ensure that the pixel values ​​of the output image are within a reasonable range. The reconstructed image is the original image I I It learns the high-level feature representation of the data by optimizing the reconstruction error, where the reconstruction error is usually calculated using the mean square error (MSE). Through the decoding process of the autoencoder, the model can learn how to reconstruct images from low-dimensional feature vectors, which helps the model use these feature representations for more complex data analysis and processing in subsequent tasks.

[0075] In step S304, after feature selection and dimensionality reduction processing: after the autoencoder training is completed, feature selection is performed on the output of the encoding layer to determine which features are most critical for cutting surface evaluation. Feature selection can be achieved by calculating the mutual information between the features and the target variable, and selecting the features with the largest mutual information. These features can distinguish different cutting surface states to the greatest extent. Finally, the feature representation extracted by deep learning for subsequent cutting surface evaluation is obtained: these features will be used as input to the cutting surface evaluation model for classification and prediction of the cutting surface state of the automobile caliper. Through feature selection and dimensionality reduction, the model can focus on the most useful features, thereby improving the accuracy and efficiency of cutting surface evaluation.

[0076] S400, obtaining a comprehensive feature representation through feature fusion processing based on the first texture feature, the second texture feature and the feature representation extracted by deep learning, wherein the feature fusion processing includes weighted averaging, PCA dimensionality reduction and Dense layer fusion method of deep learning.

[0077] It can be understood that the step S400 includes S401, S402 and S403, wherein:

[0078] S401, combining the first texture feature and the second texture feature to obtain a first physical property, the calculation formula of which is as follows: In the formula, is the fused feature set, F 1 is the first texture feature, F 2 is the second texture feature, Concat (F1, F2) is the first texture feature set F 1 and the second texture feature set F 2 are merged into a feature set, Concat represents the concatenation operation;

[0079] S402, the low-dimensional features obtained by the autoencoder and the first physical characteristic are fused, and the calculation formula is as follows: In the formula, is the final feature representation, Z is the feature extracted by deep learning, is the fused feature set, Concat represents the concatenation operation; S403, performing feature normalization, feature selection and dimension reduction on the final feature representation to obtain an optimized and comprehensive feature representation.

[0080] It should be noted that in terms of feature fusion, it includes weighted averaging and PCA dimensionality reduction, where weighted averaging is to weighted average the first texture feature and the second texture feature to fuse the two features, and PCA dimensionality reduction is to perform PCA dimensionality reduction on the features after weighted averaging to reduce the dimension of the features and extract the most useful information. In addition, Dense layer fusion, which fuses the features after PCA dimensionality reduction with the feature Z extracted by deep learning through a Dense layer (fully connected layer).

[0081] Therefore, through feature fusion processing, a comprehensive feature representation that integrates multiple feature sources can be obtained, which will greatly improve the accuracy and robustness of cutting surface evaluation and ensure that the cutting surface features captured from different angles are fully utilized, thereby providing more reliable decision support for the maintenance and replacement of automobile calipers.

[0082] S500. Processing the comprehensive feature representation through feature selection based on mutual information to obtain the feature that contributes most to the judgment of the cutting surface state, wherein the feature selection process includes calculating the mutual information between the feature and the cutting surface state, and selecting the feature with the largest mutual information as an input item.

[0083] It should be noted that "feature selection based on mutual information" is a filtering feature selection method, which evaluates the importance of features by measuring the mutual information between features and target variables. Mutual information can be understood as the amount of information a random variable contains about another random variable. The larger its value, the stronger the dependence between the two variables. In feature selection, if the mutual information between a feature and the target variable is large, it means that the feature has a higher value for the prediction of the target variable. Therefore, by calculating the mutual information between each feature and the target variable, the feature subset that contributes most to the model's prediction performance can be selected. In feature selection, if the mutual information between a feature and the target variable is large, it means that the feature has a higher value for the prediction of the target variable. Therefore, by calculating the mutual information between each feature and the target variable, the feature subset that contributes most to the model's prediction performance can be selected. Among them, the steps of mutual information feature selection are as follows:

[0084] 1. For each feature in the data set, calculate the mutual information between it and the target variable. The mutual information is calculated by the following formula: In the formula, I ( X ; Y ) is the mutual information, X is the characteristic variable, Y is the target variable, p ( x , y ) is the joint probability distribution, p (x ) and p ( y ) are marginal probability distributions, x is a random variable X A specific value of , y is a random variable Y A specific value of

[0085] 2. Sort all features according to the calculated mutual information; select the top N features with the highest mutual information as the final feature subset. These features are considered to have the strongest dependence on the target variable and contribute the most to the model's predictive performance;

[0086] 3. Train the machine learning model using the selected feature subset and train the machine learning model using the selected feature subset;

[0087] In this way, high-quality features that have significant predictive power for the target variable can be identified and selected, thereby improving the predictive accuracy and explanatory power of the model.

[0088] It can be understood that in this step, let X be the feature set extracted from the comprehensive feature representation, let Y be the cutting surface state label of the automobile caliper, and determine each feature x ∈ X and cutting surface condition y ∈ Y The probability of simultaneous occurrence is calculated, and the probability of each feature appearing alone and the probability of each cutting surface state appearing alone are calculated, and the mutual information between the feature and the cutting surface state is calculated;

[0089] For feature sets X Each feature in x , calculate its relationship with the cutting surface state Y The mutual information between I ( x ; Y ), sort the features according to the amount of mutual information, and select the feature with the largest mutual information as the feature that contributes most to the judgment of the cutting surface state.

[0090] S600. Based on the most contributing features, the classification results of the cutting surface status are obtained through long short-term memory convolutional neural network model training and optimization processing, wherein the model training and optimization include the use of cross-validation, adjustment of the network structure and selection of the optimization algorithm, wherein the classification results of the cutting surface status are used to evaluate the condition of the cutting surface of the automobile caliper.

[0091] It can be understood that in this step, a cross-validation method, such as k-fold cross-validation, is used to divide the data set into k subsets, k-1 subsets are used for training each time, and the remaining subset is used for verification, and the cycle is repeated k times, and a different subset is selected as the verification set each time; the structure of the long short-term memory convolutional neural network model is adjusted, including the number and parameters of convolutional layers, pooling layers, LSTM layers, and fully connected layers to optimize the model performance; a suitable optimization algorithm is selected, such as stochastic gradient descent (SGD), Adam, RMSprop, etc. These algorithms update the network weights in different ways to minimize the loss function; the trained and optimized model will classify the cutting surface state of the automobile caliper cutting surface, and output the classification results of the cutting surface state, which reflect the degree of the caliper cutting surface.

[0092] Embodiment 2:

[0093] like Figure 2 As shown, this embodiment provides a cutting quality assessment system for an automobile caliper production line, see Figure 2 The system comprises:

[0094] Acquisition module 701: used to collect image data of the cutting surface of the automobile caliper at different use stages, and the image data of each stage includes image information of the glossiness, color uniformity, and surface scratches of the cutting surface;

[0095] Extraction module 702: used to process the image data using a Gabor filter, extract the first texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution feature of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, and extract the second texture feature of the cutting surface of the automobile caliper, wherein the second texture feature is a Gabor filter of different scales, and each scale is designed to capture the texture details of the cutting surface within a frequency range;

[0096] Construction module 703: for constructing and training an autoencoder model based on the first texture feature and the second texture feature, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing reconstruction errors, wherein the encoder compresses high-dimensional image data into low-dimensional feature representations, and the decoder reconstructs the original image from these low-dimensional features, thereby providing deep learning extracted feature representations for cutting surface evaluation of the cutting surface of the automobile caliper;

[0097] The first processing module 704 is used to obtain a comprehensive feature representation through feature fusion processing according to the first texture feature, the second texture feature and the feature representation extracted by deep learning, wherein the feature fusion processing includes weighted averaging, PCA dimensionality reduction and Dense layer fusion method of deep learning;

[0098] The second processing module 705 is used to process the comprehensive feature representation through feature selection based on mutual information to obtain the feature that contributes most to the judgment of the cutting surface state, wherein the feature selection process includes calculating the mutual information between the feature and the cutting surface state, and selecting the feature with the largest mutual information as an input item;

[0099] Optimization module 706: used to obtain the classification result of the cutting surface state based on the most contributing features after long short-term memory convolutional neural network model training and optimization processing, wherein the model training and optimization include the use of cross-validation, adjustment of network structure and selection of optimization algorithm, wherein the classification result of the cutting surface state is used to evaluate the condition of the cutting surface of the automobile caliper.

[0100] Specifically, the extraction module 702 includes:

[0101] The first design unit is used to design Gabor filters on the cutting surface of the car caliper in different directions, where the different directions include 0°, 45°, 90° and 135°. The calculation formula is as follows: Where G is the filter, x 'and y ′ is the coordinate after rotation, l is the wavelength, i is the direction of the filter, ψ is the phase shift, s is the standard deviation of the Gaussian kernel, c is the spatial aspect ratio;

[0102] Application unit: used for applying the designed Gabor filter to the image data of the surface of the automobile caliper, wherein for the filter in each direction, a convolution operation is performed;

[0103] Extraction unit: used to extract the features of roughness, granularity and crack depth from the filtered image based on the convolution operation results.

[0104] Specifically, the building block 703 includes:

[0105] The second design unit is used to design an autoencoder model including an encoder and a decoder, wherein the encoder is composed of multiple convolutional layers and pooling layers for extracting image features, and the decoder is composed of the transposition operation of the convolutional layer for reconstructing the image;

[0106] A conversion unit: used to compress the high-dimensional image data into a low-dimensional feature vector through processing by an autoencoder model, wherein the low-dimensional feature vector contains the main information of the image; a first processing unit: used to obtain a reconstructed image based on the low-dimensional feature vector through processing by a decoder, wherein the reconstructed image is an approximate representation of the original image data, and automatically learn the high-level feature representation of the data by optimizing the reconstruction error, wherein the optimized reconstruction error includes the difference between the original image and the reconstructed image;

[0107] The second processing unit is used to obtain a low-dimensional feature representation that can represent the key information of the image based on the training results of the autoencoder model through feature selection and dimensionality reduction processing, and finally obtain a feature representation extracted by deep learning for subsequent cutting surface evaluation.

[0108] Specifically, the first processing module 704 includes:

[0109] Combining unit: used to combine the first texture feature and the second texture feature to obtain a first physical property, and the calculation formula is as follows: In the formula, is the fused feature set, F 1 is the first texture feature, F 2 is the second texture feature, Concat (F1, F2) is the first texture feature set F 1 and the second texture feature set F 2 are merged into a feature set, Concat represents the concatenation operation;

[0110] Fusion unit: used to fuse the low-dimensional features obtained by the autoencoder with the first physical characteristic. The calculation formula is as follows: In the formula, is the final feature representation, Z is the feature extracted by deep learning, is the fused feature set, Concat represents the concatenation operation; Representation unit: used to perform feature normalization, feature selection and dimensionality reduction on the final feature representation to obtain an optimized and comprehensive feature representation.

[0111] Specifically, the second processing module 705 includes:

[0112] A computing unit: used to let X be a feature set extracted from the comprehensive feature representation, let Y be a cutting surface state label of the automobile caliper, determine the probability of each feature and cutting surface state appearing at the same time, calculate the probability of each feature appearing alone and the probability of each cutting surface state appearing alone, and calculate the mutual information between the feature and the cutting surface state;

[0113] Sorting unit: used to evaluate the mutual information of each feature, sort the features, and select the feature with the largest mutual information as the feature that contributes most to the judgment of the cutting surface state.

[0114] It should be noted that, regarding the system in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0115] Embodiment 3:

[0116] Corresponding to the above method embodiment, this embodiment also provides a cutting processing quality assessment device for an automobile caliper production line. The cutting processing quality assessment device for an automobile caliper production line described below and the cutting processing quality assessment method for an automobile caliper production line described above can be referenced to each other.

[0117] Figure 3 FIG. 8 is a block diagram of a cutting quality evaluation device 800 for an automobile caliper production line according to an exemplary embodiment. Figure 3 As shown, the cutting quality assessment device 800 for the automobile caliper production line includes: a processor 801 and a memory 802. The cutting quality assessment device 800 for the automobile caliper production line also includes one or more of a multimedia component 803, an I / O interface 804, and a communication component 805.

[0118] The processor 801 is used to control the overall operation of the cutting quality assessment device 800 for the automobile caliper production line to complete all or part of the steps in the cutting quality assessment method for the automobile caliper production line. The memory 802 is used to store various types of data to support the operation of the cutting quality assessment device 800 for the automobile caliper production line, and these data may include, for example, instructions for any application or method operating on the cutting quality assessment device 800 for the automobile caliper production line, and application-related data, such as contact data, sent and received messages, pictures, audio, video, etc. The memory 802 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk. The multimedia component 803 may include a screen and an audio component. The screen may be, for example, a touch screen, and the audio component is used to output and / or input audio signals. For example, the audio component may include a microphone for receiving external audio signals. The received audio signal may be further stored in the memory 802 or sent via the communication component 805. The audio component also includes at least one speaker for outputting audio signals. The I / O interface 804 provides an interface between the processor 801 and other interface modules, and the above-mentioned other interface modules can be a keyboard, a mouse or a button, etc. These buttons can be virtual buttons or physical buttons. The communication component 805 is used for wired or wireless communication between the cutting quality assessment device 800 of the automobile caliper production line and other devices. Wireless communication, such as Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G or 4G, or a combination of one or more of them, so the corresponding communication component 805 can include: a Wi-Fi module, a Bluetooth module or an NFC module.

[0119] In an exemplary embodiment, the cutting processing quality assessment device 800 for the automobile caliper production line can be implemented by one or more application specific integrated circuits (Application Specific Integrated Circuit, referred to as ASIC), digital signal processors (Digital Signal Processor, referred to as DSP), digital signal processing devices (Digital Signal Processing Device, referred to as DSPD), programmable logic devices (Programmable Logic Device, referred to as PLD), field programmable gate arrays (Field Programmable Gate Array, referred to as FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the above-mentioned cutting processing quality assessment method for the automobile caliper production line.

[0120] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, and when the program instructions are executed by a processor, the steps of the cutting quality assessment method for the automobile caliper production line are implemented. For example, the computer-readable storage medium may be the memory 802 including the program instructions, and the program instructions may be executed by the processor 801 of the cutting quality assessment device 800 for the automobile caliper production line to complete the cutting quality assessment method for the automobile caliper production line.

[0121] Embodiment 4:

[0122] Corresponding to the above method embodiment, a readable storage medium is also provided in this embodiment. The readable storage medium described below and the cutting quality assessment method for an automobile caliper production line described above can refer to each other.

[0123] A computer program is stored on the readable storage medium, and when the computer program is executed by the processor, the steps of the cutting quality assessment method of the automobile caliper production line in the above method embodiment are implemented.

[0124] The readable storage medium may specifically be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or other readable storage medium that can store program codes.

[0125] In summary, the present invention provides a cutting quality assessment method and system for an automobile caliper production line, aiming to improve the efficiency and accuracy of cutting surface assessment through automation and intelligent technology. The method uses advanced image processing technology and deep learning algorithms to automatically extract the texture features of the cutting surface of the automobile caliper and evaluate its cutting surface state. The cutting surface degree of the cutting surface is classified through a long short-term memory convolutional neural network model, thereby achieving a rapid and accurate assessment of the cutting surface of the automobile caliper. The system can monitor and evaluate the cutting surface condition in real time, provide a scientific basis for the production and quality control of automobile calipers, and ensure the safety and reliability of automobiles.

[0126] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

[0127] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for evaluating the cutting quality of an automobile caliper production line, characterized in that: include: Collect image data of the cutting surface of the automobile caliper at different stages of use. The image data at each stage includes image information of the glossiness, color uniformity, and surface scratches of the cutting surface; The image data is processed using a Gabor filter to extract the first texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution feature of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, and the second texture feature of the cutting surface of the automobile caliper is extracted, wherein the second texture feature is a Gabor filter of different scales, and each scale is designed to capture the texture details of the cutting surface within a frequency range; Based on the first texture feature and the second texture feature, an autoencoder model is constructed and trained, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing reconstruction errors, wherein the encoder compresses high-dimensional image data into low-dimensional feature representations, and the decoder reconstructs the original image from these low-dimensional features, thereby providing deep learning extracted feature representations for evaluation of the cutting surface of the automobile caliper; According to the first texture feature, the second texture feature and the feature representation extracted by deep learning, a comprehensive feature representation is obtained through feature fusion processing, wherein the feature fusion processing includes weighted average, PCA dimension reduction and Dense layer fusion method of deep learning; Processing the comprehensive feature representation through feature selection based on mutual information to obtain the feature that contributes most to the judgment of the surface state of the cutting surface, wherein the feature selection process includes calculating the mutual information between the feature and the surface state of the cutting surface, and selecting the feature with the largest mutual information as an input item; Based on the most contributing features, the classification results of the cutting surface state are obtained through long short-term memory convolutional neural network model training and optimization processing, where model training and optimization include the use of cross-validation, adjustment of network structure and selection of optimization algorithm; Roughness: Calculate the local variance of the filtered image to assess surface irregularities: In the formula, is the average intensity of the filtered image, and N is the total number of pixels; is the filtered image; Granularity: It is evaluated by analyzing the contrast and uniformity of the gray level co-occurrence matrix GLCM of the filtered image: Granularity = GLCM (I′, θ) where GLCM calculates the co-occurrence frequency of pixel pairs in the direction θ; Crack Depth: Estimate the depth and width of cracks by detecting high-frequency components in the filtered image: In the formula, Δ x and Δ y is the difference operator in the x and y directions.

2. The cutting quality assessment method for automobile caliper production line according to claim 1 is characterized in that: The Gabor filter is used to process the image data to extract the first texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution characteristics of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, including: Design Gabor filters for the cutting surface of the car caliper in different directions, including 0°, 45°, 90° and 135° directions. The calculation formula is as follows: Where G is the filter, x 'and y ′ is the coordinate after rotation, λ is the wavelength, θ is the direction of the filter, ψ is the phase shift, σ is the standard deviation of the Gaussian kernel, γ is the spatial aspect ratio; The designed Gabor filter is applied to the image data of the car caliper surface, where a convolution operation is performed for the filter in each direction; Based on the convolution operation results, the features of roughness, granularity and crack depth are extracted from the filtered image.

3. The cutting quality assessment method for automobile caliper production line according to claim 2, characterized in that: The calculation formula for performing the convolution operation is as follows: In the formula, I ′( x , y ) is the filtered image, I ( x , y ) is the original image in spatial coordinates ( x , y ), G ( x , y , λ , θ , ψ , σ , γ ) is a complex function, which is the functional expression of Gabor filter.

4. The cutting quality assessment method for automobile caliper production line according to claim 1, characterized in that: The autoencoder model is constructed and trained, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing the reconstruction error, wherein the encoder compresses the high-dimensional image data into low-dimensional feature representations, and the decoder reconstructs the original image from these low-dimensional features, thereby providing deep learning extracted feature representations for the evaluation of the cutting surface of the automobile caliper, including: Design an autoencoder model that includes an encoder and a decoder, where the encoder consists of multiple convolutional layers and pooling layers to extract image features, and the decoder consists of the transpose operation of the convolutional layer to reconstruct the image; After being processed by the autoencoder model, the high-dimensional image data is compressed into a low-dimensional feature vector, where the low-dimensional feature vector contains the main information of the image; According to the low-dimensional feature vector, after being processed by a decoder, a reconstructed image is obtained, wherein the reconstructed image is an approximate representation of the original image data, and a high-level feature representation of the data is automatically learned by optimizing the reconstruction error, wherein the optimized reconstruction error includes the difference between the original image and the reconstructed image; According to the training results of the autoencoder model, after feature selection and dimensionality reduction processing, a low-dimensional feature representation that can represent the key information of the image is obtained, and finally a feature representation extracted by deep learning for cutting surface evaluation is obtained.

5. The cutting quality assessment method for automobile caliper production line according to claim 1, characterized in that: The feature representation extracted according to the first texture feature, the second texture feature and deep learning is subjected to feature fusion processing to obtain a comprehensive feature representation, which includes: The first texture feature and the second texture feature are combined to obtain a first physical property, which is calculated as follows: In the formula, is the fused feature set, F 1 is the first texture feature, F 2 is the second texture feature, Concat (F1, F2) is the first texture feature set F 1 and the second texture feature set F 2 are merged into a feature set, Concat represents the concatenation operation; The low-dimensional features obtained by the autoencoder are fused with the first physical characteristic, and the calculation formula is as follows: In the formula, is the final feature representation, Z is the feature extracted by deep learning, is the fused feature set, Concat represents the concatenation operation; the final feature representation is subjected to feature normalization, feature selection and dimensionality reduction to obtain an optimized and comprehensive feature representation.

6. The cutting quality assessment method for automobile caliper production line according to claim 1, characterized in that: The comprehensive feature representation is processed by feature selection based on mutual information to obtain the feature that contributes most to the judgment of the surface state of the cutting surface, wherein the feature selection process includes calculating the mutual information between the feature and the surface state of the cutting surface, and selecting the feature with the largest mutual information as an input item, including: Let X be the set of features extracted from the comprehensive feature representation, let Y be the cutting surface state label of the automobile caliper, determine the probability of each feature and the cutting surface state appearing at the same time, calculate the probability of each feature appearing alone and the probability of each cutting surface state appearing alone, and calculate the mutual information between the feature and the cutting surface state; The mutual information of each feature is evaluated, and the features are sorted. The feature with the largest mutual information is selected as the feature that contributes most to the judgment of the surface state of the cutting surface.

7. A cutting quality assessment system for an automobile caliper production line, based on the cutting quality assessment method for an automobile caliper production line according to claim 1, characterized in that: include: Acquisition module: used to collect image data of the cutting surface of the automobile caliper at different use stages. The image data at each stage includes the image information of the glossiness, color uniformity, and surface scratches of the cutting surface; Extraction module: used to process the image data using Gabor filters, extract the first texture feature of the cutting surface of the automobile caliper, wherein the first texture feature is the distribution feature of the roughness, granularity and crack depth of the cutting surface in the directions of 0°, 45°, 90° and 135°, and extract the second texture feature of the cutting surface of the automobile caliper, wherein the second texture feature is Gabor filters of different scales, and each scale is designed to capture the texture details of the cutting surface within a frequency range; A construction module: used for constructing and training an autoencoder model based on the first texture feature and the second texture feature, wherein the autoencoder model automatically extracts high-level abstract features of the data by minimizing the reconstruction error, wherein the encoder compresses the high-dimensional image data into a low-dimensional feature representation, and the decoder reconstructs the original image from the low-dimensional features, thereby providing a feature representation extracted by deep learning for the evaluation of the cutting surface of the automobile caliper; The first processing module is used to obtain a comprehensive feature representation through feature fusion processing based on the first texture feature, the second texture feature and the feature representation extracted by deep learning, wherein the feature fusion processing includes weighted averaging, PCA dimensionality reduction and Dense layer fusion method of deep learning; The second processing module is used to process the comprehensive feature representation through feature selection based on mutual information to obtain the feature that contributes most to the judgment of the surface state of the cutting surface, wherein the feature selection process includes calculating the mutual information between the feature and the surface state of the cutting surface, and selecting the feature with the largest mutual information as an input item; Optimization module: It is used to obtain the classification results of the cutting surface state based on the most contributing features after long short-term memory convolutional neural network model training and optimization processing. The model training and optimization include the use of cross-validation, adjustment of network structure and selection of optimization algorithm. The classification results of the cutting surface state are used to evaluate the cutting surface of automobile calipers.

8. The cutting quality evaluation system for automobile caliper production line according to claim 7, characterized in that: The extraction module includes: The first design unit is used to design Gabor filters on the cutting surface of the car caliper in different directions, where the different directions include 0°, 45°, 90° and 135°. The calculation formula is as follows: Where G is the filter, x 'and y ′ is the coordinate after rotation, λ is the wavelength, θ is the direction of the filter, ψ is the phase shift, σ is the standard deviation of the Gaussian kernel, γ is the spatial aspect ratio; Application unit: used for applying the designed Gabor filter to the image data of the surface of the automobile caliper, wherein for the filter in each direction, a convolution operation is performed; Extraction unit: used to extract the features of roughness, granularity and crack depth from the filtered image based on the convolution operation results.

9. The cutting quality evaluation system for automobile caliper production line according to claim 7, characterized in that: The building blocks include: The second design unit is used to design an autoencoder model including an encoder and a decoder, wherein the encoder is composed of multiple convolutional layers and pooling layers for extracting image features, and the decoder is composed of the transposition operation of the convolutional layer for reconstructing the image; A conversion unit: used to compress the high-dimensional image data into a low-dimensional feature vector through processing by an autoencoder model, wherein the low-dimensional feature vector contains the main information of the image; a first processing unit: used to obtain a reconstructed image based on the low-dimensional feature vector through processing by a decoder, wherein the reconstructed image is an approximate representation of the original image data, and automatically learn the high-level feature representation of the data by optimizing the reconstruction error, wherein the optimized reconstruction error includes the difference between the original image and the reconstructed image; The second processing unit is used to obtain a low-dimensional feature representation that can represent the key information of the image based on the training results of the autoencoder model through feature selection and dimensionality reduction processing, and finally obtain a feature representation extracted by deep learning for cutting surface evaluation.

10. The cutting quality evaluation system for automobile caliper production line according to claim 7, characterized in that: The first processing module includes: Combining unit: used to combine the first texture feature and the second texture feature to obtain a first physical property, and the calculation formula is as follows: In the formula, is the fused feature set, F 1 is the first texture feature, F 2 is the second texture feature, Concat (F1, F2) is the first texture feature set F 1 and the second texture feature set F 2 are merged into a feature set, Concat represents the concatenation operation; Fusion unit: used to fuse the low-dimensional features obtained by the autoencoder with the first physical characteristic. The calculation formula is as follows: In the formula, is the final feature representation, Z is the feature extracted by deep learning, is the fused feature set, Concat represents the concatenation operation; Representation unit: used to perform feature normalization, feature selection and dimensionality reduction on the final feature representation to obtain an optimized and comprehensive feature representation.

Images