Intelligent detection method and system for workpiece surface defects
Through intelligent detection methods, the workpiece surface defect detection is detected using end-to-end noise-de-network model and deep learning model, which solves the problem that traditional methods are difficult to achieve high-precision detection in complex industrial environments, and achieves efficient, stable and adaptive detection. Effect.
Patent Information
- Application Number
- CN202411454465.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-17
- Publication Date
- 2025-02-07
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Workpiece surface defect detection faces complex industrial environment, diverse workpiece characteristics and variable defect types, making it difficult for traditional methods to achieve high-precision, stability and adaptability detection.
An intelligent detection method is adopted, including real-time image acquisition, pre-processing to remove noise through end-to-end denoising network model, feature extraction deep learning model extracts deep features, image surface defect recognition network model identifys defect type and severity, and performs quality Evaluate.
It improves the accuracy and stability of workpiece surface defect detection, can adapt to diverse workpiece characteristics and variable defect types, and significantly improves the quality control capabilities and production efficiency of the production line.
Smart Images
Figure CN119399147A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of workpiece defect detection, and in particular to an intelligent detection method and system for workpiece surface defects. Background Art
[0002] In industrial production, workpiece surface defect detection faces many challenges, which mainly come from the harsh industrial environment and the complexity of workpiece characteristics. First, industrial environmental conditions such as high temperature, high humidity, vibration and unstable lighting will affect the performance of sensors and cameras, resulting in unstable image quality. In addition, impurities such as dust and oil adhere to the surface of the workpiece, which may cover real defects or form pseudo-defects, increasing the difficulty of detection. On high-speed production lines, the relative movement of the workpiece and the camera can easily lead to motion blur, which significantly reduces the accuracy of defect identification. At the same time, the complex texture, irregular shape and diverse reflective properties of the workpiece surface also make it difficult for traditional detection methods to effectively deal with it. In addition, the diversity and subtle differences in defect types further increase the complexity and uncertainty of detection.
[0003] In summary, the current workpiece surface defect detection faces multiple challenges such as complex industrial environment, diverse workpiece characteristics and variable defect types. Traditional methods are difficult to deal with these problems. Therefore, it is urgent to develop an intelligent detection solution that can adapt to harsh industrial environments and handle diverse workpiece surface characteristics and defect types, so as to greatly improve the accuracy, stability and adaptability of detection and meet the strict quality requirements of modern industrial production. Summary of the invention
[0004] The purpose of the present invention is to address the deficiencies in the above-mentioned background technology and provide a solution that can adapt to harsh industrial environments, handle diverse workpiece surface features and defect types, and significantly improve the accuracy, stability and adaptability of detection.
[0005] In order to achieve the above object, the present invention provides an intelligent detection method for workpiece surface defects, comprising the following steps:
[0006] S1, real-time image acquisition of workpieces conveyed on the production line;
[0007] S2, preprocessing the collected image data through a pre-trained end-to-end denoising network model;
[0008] S3, inputting the new workpiece image after denoising and restoration into the feature extraction deep learning model, and the feature extraction deep learning model conducts in-depth analysis on the workpiece image to extract the deep features of its surface;
[0009] S4, inputting the deep features into the surface defect recognition network model, classifying and processing the deep features through the surface defect recognition network model, and identifying various defect types on the workpiece surface;
[0010] S5, evaluates the overall quality of each workpiece based on the type, number, and severity of defects identified online.
[0011] Furthermore, in S2, the input workpiece image is first converted into spectrum data by fast Fourier transform:
[0012]
[0013] Where f(x,y) represents the image pixel value in the spatial domain, F(u,υ) represents the spectrum data in the frequency domain, and M and N are the width and height of the image respectively;
[0014] The spectral data is then denoised using an end-to-end denoising network model. After denoising, the spectral data is converted back into an image in the spatial domain through inverse Fourier transform to generate a denoised artifact image:
[0015]
[0016] Furthermore, the offline training of the end-to-end denoising network model in S2 includes the following steps:
[0017] S21, build an end-to-end denoising network model integrating self-attention mechanism;
[0018] S22, sorting the acquired noisy and noise-free workpiece image spectrum data to form a first training data set and inputting the first training data set into an end-to-end denoising network model integrating a self-attention mechanism;
[0019] S23, set key hyperparameters, including: training batch size, to determine the amount of data involved in the calculation in each training; learning rate, used to control the update amplitude of the model; maximum number of iterations, to determine the maximum number of cycles in the training process;
[0020] S24, offline training of the end-to-end denoising network model.
[0021] Furthermore, the feature extraction deep learning model in S3 gradually extracts high-dimensional features in the image through layer-by-layer feature extraction.
[0022] Furthermore, the image surface defect recognition network model in S4 classifies each defect into a preset category through feature learning and labels the severity level for each defect category.
[0023] Furthermore, the offline training of the image surface defect recognition network model in S4 includes the following steps:
[0024] S41, constructing a network model for image surface defect recognition that integrates coordinated attention mechanism;
[0025] S42, preparing a second training data set: collecting and sorting the workpiece surface images without noise; matching the manually annotated defect types and their severity levels with the image data; inputting the sorted data set as the second training data set into the constructed image surface defect recognition network model with fusion coordinated attention mechanism;
[0026] S43, set the necessary hyperparameters, including: the batch size of the training, that is, the number of samples used in each training; the learning rate, which is used to control the rate at which the model weights are updated; the maximum number of iterations to ensure that the training process is completed in a reasonable time;
[0027] S44, offline training of the image surface defect recognition network model.
[0028] Furthermore, in S5, based on pre-set evaluation criteria, workpieces are divided into three levels: qualified products, defective products and scrap products; if the number and severity of defects in the workpiece are within an acceptable range, it is evaluated as a qualified product; if the defects exceed the preset number but are still within the repairable or usable standard, it is evaluated as a defective product; and when the defects are too many or serious, it is evaluated as scrap products.
[0029] Furthermore, S5 includes the following sub-steps:
[0030] S51, collecting images of workpieces to be inspected on the production line;
[0031] S52, converting the workpiece image into spectrum data by fast Fourier transform;
[0032] S53, inputting the converted spectrum data into a trained end-to-end denoising network model for denoising, to obtain denoised workpiece image spectrum data;
[0033] S54, performing inverse Fourier transform on the denoised workpiece image spectrum data to restore it into a new workpiece image, first inputting the feature extraction deep learning model to extract deep features, and then inputting the deep features into the trained image surface defect recognition network model;
[0034] S55, the image surface defect recognition network model classifies the types and severity of defects and outputs the recognition results of various types of defects and severity levels;
[0035] S56, evaluating the identified defects and severity levels, and calculating scores based on the number of defects and the corresponding severity levels;
[0036] S57, the workpieces evaluated as qualified products are directly sent to the packaging area through the conveyor belt, the workpieces evaluated as defective products are reprocessed for unqualified parts, and the workpieces evaluated as waste are directly scrapped.
[0037] Furthermore, in S56, the basic scores corresponding to the four types of defects are set to be x1, x2, x3 and x4, respectively, and the scores corresponding to severity levels 1 to 7 are a1, a2, ..., a7, respectively. The scoring rules for each defect detection result on a single workpiece are as follows:
[0038] xc i =x i +a j
[0039] Among them, xc i is the score of the i-th defect, and j is the severity level;
[0040] Get the overall score xc of a single artifact:
[0041]
[0042] Where N is the total number of defect types;
[0043] Qualified products, defective products and waste products are classified according to the following rules:
[0044]
[0045] The present invention also provides an intelligent detection system for workpiece surface defects, which adopts the intelligent detection method for workpiece surface defects as described above, including a workpiece surface image acquisition module, an image preprocessing module, a surface feature extraction module, an image surface defect recognition module and a quality assessment module;
[0046] The workpiece surface image acquisition module includes an industrial camera and a fixed bracket. The industrial camera is supported at a preset position of the conveyor belt of the production line by the fixed bracket, and is used to perform real-time image acquisition of the workpiece on the conveyor belt of the production line, and send the acquired image data to the image preprocessing module; the image preprocessing module first converts the input workpiece image into a spectrum by fast Fourier transform, and then inputs the spectrum data into the trained end-to-end denoising network model for denoising, outputs the denoised spectrum, and the denoised spectrum is restored to a new workpiece image by inverse Fourier transform, and the new workpiece image after denoising and restoration is input into the surface feature extraction module; the surface feature extraction module extracts the deep features of the workpiece image by a feature extraction deep learning model, and inputs the deep features into the image surface defect recognition module; the image surface defect recognition module classifies the input deep features of the image by the image surface defect recognition network model, and outputs the recognition results and severity levels of various defects of the workpiece to the quality assessment module; the quality assessment module evaluates the production quality of lithium battery pole pieces according to the number of recognition results and severity levels of the defects.
[0047] The above scheme of the present invention has the following beneficial effects:
[0048] The intelligent detection method and system for workpiece surface defects provided by the present invention effectively eliminate environmental noise through an end-to-end denoising network model, improve the clarity and accuracy of the image, and thus improve the accuracy of lithium battery pole piece surface defect detection; through the combination of a feature extraction deep learning model and an image surface defect recognition network model, the identified defects can be classified to ensure that only qualified products are output during the production process, effectively improving the overall quality and safety of lithium battery pole pieces; the method can process a large amount of production data in real time, realize rapid classification and processing of lithium battery pole pieces, and significantly improve the efficiency of the production line;
[0049] Other beneficial effects of the present invention will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 is a flow chart of the steps of the present invention;
[0051] Figure 2 It is a flowchart of the operation of offline training of the end-to-end denoising network model of the present invention;
[0052] Figure 3 It is a schematic diagram of building an end-to-end denoising network model integrating the self-attention mechanism;
[0053] Figure 4 It is the operation flow chart of offline training of the network model for image surface defect recognition;
[0054] Figure 5It is a schematic diagram of constructing a network model for image surface defect recognition that integrates coordinated attention mechanism;
[0055] Figure 6 It is a detailed step flow chart of S5;
[0056] Figure 7 It is a schematic diagram of the workpiece surface image acquisition module;
[0057] Figure 8 This is the operation flow chart of the image preprocessing module based on frequency domain noise reduction;
[0058] Fig. 9 It is the operation flow chart of the surface feature extraction module and the image surface defect recognition module;
[0059] Fig.10 It is the operation flow chart of the quality assessment module. DETAILED DESCRIPTION
[0060] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.
[0061] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.
[0062] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure, and the illustrations only show the components related to the present disclosure rather than the number, shape and size of the components in actual implementation. The type, quantity and proportion of each component in actual implementation may be changed at will, and the component layout type may also be more complicated. In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.
[0063] like Figure 1 As shown, an embodiment of the present invention provides an intelligent detection method for workpiece surface defects. Taking the production line of lithium battery pole pieces as an example, an end-to-end denoising network model is used to preprocess the lithium battery pole piece image, and the influence of environmental noise on the image quality is eliminated by spectrum conversion, thereby improving the clarity and accuracy of the image. The denoised image is then analyzed by a surface defect recognition network model to accurately identify defects such as exposed foil, dark spots, spots, and material dropouts that may exist on the surface of the lithium battery pole piece, and these defects are classified and labeled. The recognition results are then evaluated for quality, and the lithium battery pole pieces are classified into qualified products, defective products, and waste products according to the type and severity of the defects for subsequent processing on the production line. This method not only improves the accuracy and efficiency of defect detection, but also can process a large amount of production data in real time, significantly improves the quality control capability of the production line, and ensures the performance and safety of lithium battery products.
[0064] Specifically in this embodiment, the method includes the following steps:
[0065] S1, real-time image acquisition of lithium battery electrodes conveyed on the production line.
[0066] S2, preprocessing the collected image data through a pre-trained end-to-end denoising network model.
[0067] Among them, the preprocessing is based on frequency domain noise reduction. The input lithium battery pole piece image is first converted into spectrum data through fast Fourier transform:
[0068]
[0069] Among them, f(x,y) represents the image pixel value in the spatial domain, F(u,υ) represents the spectrum data in the frequency domain, and M and N are the width and height of the image respectively.
[0070] It should be noted that the distinction between noise and signal is more obvious in the frequency domain, so processing images in the frequency domain can more effectively remove noise. By introducing a pre-trained end-to-end denoising network model, the spectrum data can be denoised to remove possible noise interference. Among them, the end-to-end denoising network model is trained with a large amount of noisy and noise-free data, and can identify and remove noise in complex environments, thereby retaining effective information in the spectrum. After denoising is completed, the spectrum data is converted back into an image in the spatial domain through the inverse Fourier transform of the following formula to generate a denoised lithium battery pole piece image:
[0071]
[0072] At the same time Figure 2 As shown, the offline training of the end-to-end denoising network model in this embodiment specifically includes:
[0073] S21, build an end-to-end denoising network model integrating self-attention mechanism, such as Figure 3 As shown in the figure, the end-to-end denoising network model uses a four-layer convolutional neural network to extract image features, uses a batch normalization layer to accelerate the training process and enhance the stability of the model, and introduces a maximum pooling operator to reduce the size of the image feature map extracted by the convolutional neural network. It should be noted that the self-attention mechanism is used to focus on the correlation of each spatial position in the lithium battery pole piece image to improve the denoising effect of the model.
[0074] S22, the obtained noisy and noise-free lithium battery pole piece image spectrum data are sorted to form a first training data set and input into the constructed end-to-end denoising network model integrating the self-attention mechanism. The spectrum data in the data set is preprocessed to ensure that the data format matches the input requirements of the network model. By inputting these spectrum data, the model learns how to effectively remove noise from noisy images and keep the original spectrum information of the image.
[0075] S23, (before starting training) set key hyperparameters, including: the batch size of the training to determine the amount of data involved in the calculation in each training; the learning rate, which is used to control the update amplitude of the model to avoid excessive or slow updates; and the maximum number of iterations, which determines the maximum number of cycles in the training process, so as to ensure that the model can converge to an optimal solution in a reasonable time.
[0076] Among them, the end-to-end denoising network model uses mean square error as the loss function:
[0077]
[0078] in, is the output spectrum data of the denoising network, y iis the corresponding noise-free real spectrum data, and n is the total amount of data.
[0079] S24, offline training of the end-to-end denoising network model. By repeatedly iteratively training on a large amount of lithium battery pole piece image spectrum data, the model can gradually optimize its denoising ability and finally obtain a trained end-to-end denoising network model. This trained model can be used in subsequent online denoising tasks to help improve the quality and stability of lithium battery pole piece images.
[0080] S3, inputs the new lithium battery pole piece image after denoising and restoration into the feature extraction deep learning model, and the feature extraction deep learning model conducts in-depth analysis on the lithium battery pole piece image to extract the deep features of its surface.
[0081] Among them, the feature extraction deep learning model can gradually extract high-dimensional features in the image through layer-by-layer feature extraction, including texture, edge, brightness and other information, which can fully reflect the potential defect characteristics on the surface of lithium battery pole pieces. The extracted deep features are used as input for subsequent image surface defect recognition.
[0082] S4, inputting the deep features into the surface defect recognition network model, classifying and processing the deep features through the surface defect recognition network model, and identifying various defect types on the surface of the lithium battery electrode.
[0083] In this step, the trained image surface defect recognition network model can identify four common types of defects such as exposed foil, dark spots, spots, and material dropouts, and can also classify and output them according to the severity of the defects. Specifically, the image surface defect recognition network model classifies each defect into a specific category through feature learning, and labels the severity level of each defect, thereby providing a more detailed basis for subsequent quality assessment.
[0084] At the same time Figure 4 As shown, the offline training of the image surface defect recognition network model in this embodiment specifically includes:
[0085] S41, build an image surface defect recognition network model integrating coordinated attention mechanism, such as Figure 5 As shown in the figure, the core architecture of the image surface defect recognition network model integrating coordinated attention is based on a multi-layer residual network. The residual network uses jump connections to solve the gradient vanishing problem in the deep network, enhancing the learning ability of the deep features of the lithium battery pole piece image. The integrated coordinated attention mechanism enables the network to capture the subtle defects on the surface of the lithium battery pole piece image across a long distance while paying attention to local features, helping the model to focus on the key areas of defects while taking into account the entire image surface information, thereby improving the accuracy and robustness of recognition.
[0086] S42, prepare a second training data set for offline training of the image surface defect recognition network model. First, collect and organize the noise-free surface images of lithium battery pole pieces to ensure the quality and consistency of the image data. Then, combine the manually annotated defect types and their severity levels to match this information with the image data. These annotations include different types of surface defects such as exposed foil, dark spots, spots, and material dropouts, as well as the corresponding severity level classifications. The data set is strictly preprocessed to ensure that the model can accurately learn the defect features and the severity of various defects during the training process. Finally, the organized data set is input as the second training data set into the constructed image surface defect recognition network model with a fusion coordinated attention mechanism.
[0087] S43, (before starting training) set the necessary hyperparameters, including: batch size, that is, the number of samples used in each training; learning rate, which is used to control the rate at which the model weights are updated to balance the convergence speed and accuracy of the model; and the maximum number of iterations to ensure that the training process is completed in a reasonable time.
[0088] S44, offline training of the image surface defect recognition network model is performed, and the second training data set is repeatedly input into the model. The model continuously optimizes its recognition ability and gradually learns how to accurately identify and classify different surface defects. The addition of the coordinated attention mechanism enables the model to remain highly sensitive to subtle defects when processing a large range of surface images. During the training process, the model gradually adjusts the network weights through the loss function, and finally obtains a fully trained image surface defect recognition network model.
[0089] S5, evaluates the overall quality of each lithium battery electrode based on the defect type, number and severity identified online.
[0090] In this step, lithium battery pole pieces are divided into three levels based on pre-set evaluation criteria: qualified, defective, and scrap. If the number and severity of defects in the lithium battery pole piece are within an acceptable range, it is rated as a qualified product; if the defects exceed a certain number but are still within the standard of being repairable or usable, it is rated as a defective product; and when the defects are too many or serious, it is rated as scrap. Through this intelligent evaluation system, quality control can be carried out in real time during the production process, improving production efficiency and reducing scrap rates.
[0091] At the same time Figure 6 As shown, this step may specifically include the following sub-steps:
[0092] S51, collecting images of lithium battery pole pieces to be inspected on the production line.
[0093] S52, converting the lithium battery electrode image into spectrum data through fast Fourier transform.
[0094] S53, inputting the converted spectrum data into the trained end-to-end denoising network model for denoising, and obtaining denoised lithium battery pole piece image spectrum data.
[0095] S54, performing inverse Fourier transform on the denoised lithium battery pole piece image spectrum data to restore it into a new lithium battery pole piece image, first inputting the feature extraction deep learning model to extract deep features, and then inputting the deep features into the trained image surface defect recognition network model.
[0096] S55, the image surface defect recognition network model classifies the type and severity of the defects (through the Softmax function), and outputs the recognition results of four types of defects and severity levels, namely, exposed foil, dark spots, spots, and material drop.
[0097] S56, evaluating the identified defects and severity levels, and calculating scores based on the number of defects and the corresponding severity levels.
[0098] Specifically, the basic scores corresponding to the exposed foil, dark spots, spots and material drop defects are x1, x2, x3 and x4 respectively, and the scores corresponding to the severity levels 1 to 7 are a1, a2, ..., a7 respectively. The scoring rules for each defect detection result on a single workpiece are as follows:
[0099] xc i =x i +a j
[0100] Among them, xc i is the score of the i-th defect, and j is the severity level.
[0101] Get the overall score xc of a single artifact:
[0102]
[0103] Where N is the total number of defect types.
[0104] Qualified products, defective products and waste products are classified according to the following rules:
[0105]
[0106] S57, lithium battery electrodes that are evaluated as qualified products are directly sent to the packaging area through a conveyor belt, lithium battery electrodes that are evaluated as defective products are reprocessed for unqualified parts, and lithium battery electrodes that are evaluated as waste are directly scrapped.
[0107] As described above, the intelligent detection method for workpiece surface defects provided in this embodiment effectively eliminates environmental noise through an end-to-end denoising network model, improves the clarity and accuracy of the image, and thus improves the accuracy of surface defect detection of lithium battery pole pieces; through the combination of a feature extraction deep learning model and an image surface defect recognition network model, the identified defects can be classified to ensure that only qualified products are output during the production process, effectively improving the overall quality and safety of lithium battery pole pieces; this method can process a large amount of production data in real time, realize rapid classification and processing of lithium battery pole pieces, and significantly improve the efficiency of the production line.
[0108] Based on the same inventive concept, this embodiment also provides an intelligent detection system for workpiece surface defects, including a workpiece surface image acquisition module, an image preprocessing module, a surface feature extraction module, an image surface defect recognition module and a quality assessment module.
[0109] At the same time Figure 7 As shown, the workpiece surface image acquisition module includes an industrial camera 11 and a fixed bracket 12. The industrial camera 11 is supported at a preset position of the production line conveyor belt by the fixed bracket 12, and is used to perform real-time image acquisition of the lithium battery pole piece on the production line conveyor belt 13, and send the acquired image data to the image preprocessing module. Figure 8 As shown in the figure, the image preprocessing module first converts the input lithium battery pole piece image into a spectrum through fast Fourier transform, then inputs the spectrum data into the trained end-to-end denoising network model for denoising, outputs the denoised spectrum, and restores the denoised spectrum into a new lithium battery pole piece image through inverse Fourier transform to achieve frequency domain denoising. The new lithium battery pole piece image after denoising and restoration is input into the surface feature extraction module. Fig. 9 As shown in the figure, the surface feature extraction module extracts the deep features of the lithium battery pole piece image through the feature extraction deep learning model and inputs them into the image surface defect recognition module. The image surface defect recognition module classifies the input image deep features through the image surface defect recognition network model and outputs the recognition results and severity levels of the four types of defects of the lithium battery pole piece, namely, exposed foil, dark spots, spots, and material drop, to the quality assessment module. Fig.10 As shown, the quality assessment module is used to assess the production quality of lithium battery pole pieces according to the number of defect recognition results and the severity level, and divide them into three levels: qualified products, defective products and waste products.
[0110] The intelligent detection system for workpiece surface defects provided in this embodiment has the same inventive concept and beneficial effects as the aforementioned method, which will not be described in detail here.
[0111] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0112] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. An intelligent detection method for workpiece surface defects, characterized in that: The steps include: S1, real-time image acquisition of workpieces conveyed on the production line; S2, preprocessing the collected image data through a pre-trained end-to-end denoising network model; S3, inputting the new workpiece image after denoising and restoration into the feature extraction deep learning model, and the feature extraction deep learning model conducts in-depth analysis on the workpiece image to extract the deep features of its surface; S4, inputting the deep features into the surface defect recognition network model, classifying and processing the deep features through the surface defect recognition network model, and identifying various defect types on the workpiece surface; S5, evaluates the overall quality of each workpiece based on the type, number, and severity of defects identified online.
2. The intelligent detection method for workpiece surface defects according to claim 1, characterized in that: In S2, the input workpiece image is first converted into spectrum data through fast Fourier transform: Among them, f(x,y) represents the image pixel value in the spatial domain, F(u,v) represents the spectrum data in the frequency domain, and M and N are the width and height of the image respectively; The spectral data is then denoised using an end-to-end denoising network model. After denoising, the spectral data is converted back into an image in the spatial domain through inverse Fourier transform to generate a denoised artifact image:
3. An intelligent detection method for workpiece surface defects according to claim 1 or 2, characterized in that: The offline training of the end-to-end denoising network model in S2 includes the following steps: S21, build an end-to-end denoising network model integrating self-attention mechanism; S22, sorting the acquired noisy and noise-free workpiece image spectrum data to form a first training data set and inputting the first training data set into an end-to-end denoising network model integrating a self-attention mechanism; S23, set key hyperparameters, including: training batch size, to determine the amount of data involved in the calculation in each training; learning rate, used to control the update amplitude of the model; maximum number of iterations, to determine the maximum number of cycles in the training process; S24, offline training of the end-to-end denoising network model.
4. The intelligent detection method for workpiece surface defects according to claim 1, characterized in that: The feature extraction deep learning model in S3 gradually extracts high-dimensional features from the image through layer-by-layer feature extraction.
5. The intelligent detection method for workpiece surface defects according to claim 1, characterized in that: The image surface defect recognition network model in S4 classifies each defect into a preset category through feature learning and labels the severity level for each defect category.
6. The intelligent detection method for workpiece surface defects according to claim 5, characterized in that: The offline training of the image surface defect recognition network model in S4 includes the following steps: S41, constructing a network model for image surface defect recognition that integrates coordinated attention mechanism; S42, preparing a second training data set: collecting and arranging noise-free workpiece surface images; Combine the manually annotated defect types and their severity levels and match them with the image data; the sorted data set is used as the second training data set and input into the constructed image surface defect recognition network model with fusion coordinated attention mechanism; S43, set the necessary hyperparameters, including: the batch size of the training, that is, the number of samples used in each training; the learning rate, which is used to control the rate at which the model weights are updated; the maximum number of iterations to ensure that the training process is completed in a reasonable time; S44, offline training of the image surface defect recognition network model.
7. The intelligent detection method for workpiece surface defects according to claim 1, characterized in that: In S5, workpieces are classified into three levels based on pre-set evaluation criteria: qualified products, defective products, and scrapped products. If the number and severity of defects of a workpiece are within an acceptable range, it is rated as a qualified product. If the defects exceed the preset number but are still within the repairable or usable standards, it is considered a defective product; when the defects are too many or serious, it is considered a scrap product.
8. The intelligent detection method for workpiece surface defects according to claim 7, characterized in that: S5 includes the following sub-steps: S51, collecting images of workpieces to be inspected on the production line; S52, converting the workpiece image into spectrum data by fast Fourier transform; S53, inputting the converted spectrum data into a trained end-to-end denoising network model for denoising, to obtain denoised workpiece image spectrum data; S54, performing inverse Fourier transform on the denoised workpiece image spectrum data to restore it into a new workpiece image, first inputting the feature extraction deep learning model to extract deep features, and then inputting the deep features into the trained image surface defect recognition network model; S55, the image surface defect recognition network model classifies the types and severity of defects and outputs the recognition results of various types of defects and severity levels; S56, evaluating the identified defects and severity levels, and calculating scores based on the number of defects and the corresponding severity levels; S57, workpieces evaluated as qualified products are directly sent to the packaging area through a conveyor belt, workpieces evaluated as defective products are reprocessed at unqualified locations, and workpieces evaluated as waste are directly scrapped.
9. The intelligent detection method for workpiece surface defects according to claim 8, characterized in that: In S56, the basic scores corresponding to the four types of defects are set to be x1, x2, x3 and x4, and the scores corresponding to severity levels 1 to 7 are set to be a1, a2, ..., a7, respectively. The scoring rules for each defect detection result on a single workpiece are as follows: xc i =x i +a j Among them, xc i is the score of the i-th defect, and j is the severity level; Get the overall score xc of a single artifact: Where N is the total number of defect types; Qualified products, defective products and waste products are classified according to the following rules:
10. An intelligent detection system for workpiece surface defects, using an intelligent detection method for workpiece surface defects as claimed in any one of claims 1 to 9, characterized in that: It includes a workpiece surface image acquisition module, an image preprocessing module, a surface feature extraction module, an image surface defect recognition module and a quality assessment module; The workpiece surface image acquisition module includes an industrial camera and a fixed bracket. The industrial camera is supported at a preset position of the conveyor belt of the production line by the fixed bracket, and is used to perform real-time image acquisition of the workpiece on the conveyor belt of the production line, and send the acquired image data to the image preprocessing module; the image preprocessing module first converts the input workpiece image into a spectrum by fast Fourier transform, and then inputs the spectrum data into the trained end-to-end denoising network model for denoising, outputs the denoised spectrum, and the denoised spectrum is restored to a new workpiece image by inverse Fourier transform, and the new workpiece image after denoising and restoration is input into the surface feature extraction module; the surface feature extraction module extracts the deep features of the workpiece image by a feature extraction deep learning model, and inputs the deep features into the image surface defect recognition module; the image surface defect recognition module classifies the input deep features of the image by the image surface defect recognition network model, and outputs the recognition results and severity levels of various defects of the workpiece to the quality assessment module; the quality assessment module evaluates the production quality of lithium battery pole pieces according to the number of recognition results and severity levels of the defects.