Universal model-based modeling and detection method for defect detection in panel production lines

Abstract

A universal model-based modeling and detection method and system for defect detection in panel production lines, a device, and a storage medium. The method comprises: step 1, collecting panel production line sample data and performing defect annotation to obtain a sample data set; step 2, using the data set for modeling and training to obtain a localization model; step 3, on the basis of the data set, cropping images each containing a defect and classifying the images to form defect image data sets; step 4, using the defect image data sets for modeling and training to obtain an image classifier, and cascading the image classifier with the localization model; and step 5, on the basis of factory attributes and manufacturing process attributes, mapping defects to specific defect names by means of the image classifier and the localization model. In the method, accurate defect name mapping can be performed according to specific production environments and requirements. The overall solution reduces communication and checking time, increases sales, and also improves the applicability and accuracy of detection results.

Description

A panel production line defect detection general model modeling detection method

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] The present disclosure claims priority to the Chinese patent application No. 2024109644614, filed on July 18, 2024, and entitled "A panel production line defect detection general model modeling detection method", the entire content of which is incorporated herein by reference. TECHNICAL FIELD

[0003] The present disclosure relates to the technical field of intelligent manufacturing, in particular, to a panel production line defect detection general model modeling detection method, system, device and storage medium. BACKGROUND

[0004] In modern industrial production, panels (such as liquid crystal display screens, OLED display screens, etc.) are crucial components in electronic products. With the development of technology and the increase of market demand, the production volume and variety of panels are increasing. However, in the production process of panels, various defects such as foreign matter, residue, scratches, dirt, film breakage, and missing are inevitable. Panel defects not only reduce product quality, but also increase production costs, and even affect the stability of the entire supply chain. Therefore, panel defect detection is a key link in the production process.

[0005] Currently, panel defect detection mainly relies on manual detection and traditional detection equipment. Although manual detection is flexible, it is low in efficiency, high in cost, susceptible to human factors, and with the increase of production speed and product complexity, the limitations of manual detection become more and more significant. Traditional detection equipment usually detects based on fixed rules and features, which can provide certain detection effect in specific scenarios, but it is difficult to cope with complex and variable defect types and changes in actual production environment.

[0006] Existing automated detection systems mostly use pre-set rules and features for detection, which can easily miss and misdetect complex defect types. In addition, the process flow and product types of different factories and production lines differ, and traditional detection systems are difficult to adapt to such diverse needs. For example, when engineers check rules with customers, the rules are often not unified, resulting in low model accuracy and repeated iterations, which can cause long project delivery time and make it difficult to achieve the goal of rapid delivery in a short period. Therefore, it is particularly important to develop a defect detection system that can quickly adapt to different defect types, different production environments, and has high precision and high efficiency.

[0007] In recent years, with the development of deep learning technology, image recognition and classification technology based on deep learning has been widely applied in various fields. Deep learning has strong feature extraction ability and adaptability, can process complex image data, recognize fine image features, and has achieved remarkable results in medical image analysis, autonomous driving, intelligent security and other fields.

[0008] In the field of panel defect detection, deep learning also shows great potential. Through large-scale data training, deep learning models can automatically learn and extract defect features, thereby achieving high-precision defect detection. However, existing deep learning applications in panel defect detection still face some challenges, such as insufficient training data, weak model generalization ability, and balance between detection speed and accuracy. Therefore, how to build an efficient and universal defect detection model has become a hot and difficult research topic. SUMMARY

[0009] To solve the problem of long project delivery time ultimately caused by the background technology mentioned above, the present disclosure provides a panel production line defect detection universal model modeling detection method, system, device and storage medium. Specifically, it includes a panel production line defect detection universal model modeling detection method, which includes the following steps: Step 1, collect panel production line sample data and perform defect labeling to obtain a sample data set; beneficial effect: ensure the diversity and coverage of the data, improve the generalization ability and detection accuracy of the model. Step 2, model and train the data set to obtain a positioning model; beneficial effect: through deep learning training, obtain efficient defect positioning ability, improve the accuracy and efficiency of defect detection. Step 3, based on the data set, crop images containing defects and classify them to form a defect image data set; beneficial effect: extract and classify defect images, refine defect features, and provide accurate data basis for subsequent classifier training. Step 4, model and train the defect image data set to obtain an image classifier, and connect it with the positioning model; beneficial effect: combine positioning and classification capabilities to form a complete detection model that can simultaneously realize defect positioning and classification, improving detection efficiency. Step 5, according to the factory attributes and process attributes, map the defects to specific defect names through the image classifier and positioning model. Beneficial effect: according to the specific production environment and demand, accurately map the defect names, reduce the communication and checking time of the overall scheme, increase the sales while also improve the applicability and accuracy of the detection results.

[0010] Also disclosed are features, the panel production line sample data includes panel defect images of multiple different processes and production lines. Beneficial effect: through diversified data sources, enhance the adaptability and robustness of the model to different production environments and defect types.

[0011] Also disclosed are features that the dataset is modeled using a target detection algorithm. Advantage: using a target detection algorithm, improving the accuracy and speed of defect positioning, adapting to complex production environments.

[0012] Also disclosed are features that the target detection algorithm is a YOLO, SSD or Faster R-CNN model. Using advanced target detection algorithms ensures excellent performance of the positioning model in speed and accuracy.

[0013] Also disclosed are features that in step 5, the factory attributes and process attributes include but are not limited to process parameters, machine number, product model. By introducing various attributes, the accuracy and adaptability of defect name mapping are enhanced, which adapts to different production conditions and requirements.

[0014] Also disclosed are features that in step 5, computer vision post-processing is required before mapping. Advantage: through computer vision post-processing, the accuracy and robustness of defect mapping are improved, ensuring the reliability of the detection results.

[0015] Also disclosed are features that the computer vision post-processing includes image enhancement, feature extraction and rule-based classification algorithm. Advantage: through image enhancement and feature extraction and other processing, the image quality and feature expression ability are improved, further improving the detection and mapping effect.

[0016] And a corresponding panel production line defect detection general model modeling and detection system, comprising:

[0017] A data collection and labeling module configured to collect panel production line sample data and perform defect labeling to obtain a sample dataset, ensuring data diversity and labeling accuracy, providing a high-quality data foundation for subsequent modeling and training;

[0018] A positioning model training module configured to train a positioning model, through training to obtain an efficient positioning model, realizing accurate positioning of defects and improving detection efficiency;

[0019] A defect classification module configured to crop images containing defects and classify them to form a defect image dataset, model and train an image classifier using the defect image dataset, extract and classify defect images through the classification module, and improve the training effect and classification accuracy of the classifier;

[0020] A defect detection module configured to connect the positioning model and the image classifier in series, map the defects to specific defect names, realize positioning and classification of defects through the series connection of the positioning model and the classifier, and perform accurate mapping to improve the overall performance of the detection system.

[0021] And a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the panel production line defect detection general model modeling detection method when executing the computer program, and the detection method is implemented by the computer device to ensure efficient execution and stable operation of the method.

[0022] And a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the panel production line defect detection general model modeling detection method, and the detection method is stored and executed by the computer readable storage medium to ensure convenient deployment and wide application of the method. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings needed in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present disclosure, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0024] FIG. 1 is a flowchart of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] It should be understood that the specific embodiments described herein are only used to explain the present disclosure and not to limit the present disclosure.

[0026] To solve the defects of the prior art mentioned in the background art, the present embodiment provides a panel production line defect detection general model modeling detection method, as shown in FIG. 1, which includes the following steps:

[0027] Step 1, collect panel production line sample data and perform defect labeling to obtain a sample data set, wherein the panel production line sample data includes panel defect images of multiple different processes and production lines;

[0028] These images can be captured in real time by high-resolution cameras installed on the production line, or extracted from historical data. In order to improve the diversity of image acquisition, multi-angle, multi-spectral imaging technology can be used to increase image acquisition in different spectral ranges such as infrared and ultraviolet, and capture more defect features.

[0029] Each process may include different process stages such as coating, exposure, etching, development, etc., and each stage may produce different types of defects. By analyzing the defect distribution of each process stage, a targeted defect collection plan is developed to ensure balanced defect data for each process stage.

[0030] The collected images are classified and organized according to process, production line, time sequence, etc. properties to ensure the orderliness and traceability of the data. The database management system (such as MySQL, MongoDB, etc.) is used to manage the data, and the index is established to improve the data retrieval efficiency.

[0031] For each image, record the relevant metadata, including the collection time, equipment number, product model, etc. information. When recording metadata, a unique identifier (UUID) can be used to assign a unique identifier to each image to facilitate subsequent tracking and management.

[0032] Use image annotation tools (such as LabelImg, VGG Image Annotator, etc.) to label defects in the image. The labeling content includes the location, shape, type, etc. of the defect, ensuring the accuracy and consistency of the labeling, and the labeled data is stored in XML or JSON format for subsequent processing. In this embodiment, a double labeling mechanism is adopted, that is, two people label the same image, and finally the intersection or consistent part is taken to ensure the quality of the labeling. In order to further improve the accuracy of the labeling, an automatic labeling suggestion system based on clustering analysis can be introduced, using K-means or DBSCAN algorithm for clustering analysis of defect images to provide automatic labeling suggestions.

[0033] The labeled image data is summarized to form a complete sample data set. The sample data set should contain multiple types of defects to ensure the generalization ability of the model. During the data set construction process, a generative adversarial network (GAN) can be used to generate synthetic defect images to expand the diversity of the data set and enhance the generalization ability of the model.

[0034] Step 2, using the data set to model and train the positioning model using target detection algorithm, according to the project requirements and computing resources to select the appropriate target detection algorithm, such as YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector) or Faster R-CNN (Region-based Convolutional Neural Networks), the characteristics of each algorithm:

[0035] YOLO: good real-time performance, suitable for real-time detection.

[0036] SSD: balances detection speed and accuracy, suitable for medium-scale detection tasks.

[0037] Faster R-CNN: high detection accuracy, but large amount of calculation, suitable for high-precision demand scenarios.

[0038] Different target detection models can be customized according to the needs of different factories and processes. In model selection, transfer learning can be combined to fine-tune pre-trained models on large general-purpose datasets (such as COCO, ImageNet) to adapt to specific defect detection tasks. In addition, multi-task learning models can be designed to perform defect detection and classification simultaneously, improving the overall performance of the model.

[0039] According to different model characteristics, single selection can be used, or cross combination can be used, and a new hybrid network structure is preferentially created. The hybrid model is trained in parallel on multiple GPUs, and the loss function used combines cross-entropy loss and IoU loss to optimize the model's positioning ability for defects. In the loss function, Focal Loss can be introduced to handle the class imbalance problem of samples and improve the model's detection ability for small sample class defects.

[0040] During training, a dynamic learning rate adjustment strategy such as cosine annealing scheduling is used to improve the convergence speed and performance of the model. Optionally, the learning rate update formula is:

[0041] where η t is the current learning rate, η min and η max are the minimum and maximum learning rates, T cur is the current iteration number, and T max is the maximum iteration number.

[0042] where the sample dataset (the dataset) needs to be preprocessed, including image normalization, scaling, data augmentation (such as rotation, flipping, noise addition, etc.) to increase data diversity. Then the selected target detection model is constructed (i.e. the positioning model after construction is completed), and the preprocessed sample dataset is used for training.

[0043] During training, appropriate hyperparameters such as learning rate, batch size, iteration number, etc. are set, and the validation set is used to monitor model performance to prevent overfitting. During hyperparameter optimization, Bayesian optimization algorithm can be used to automatically search for the best hyperparameter combination to improve model performance.

[0044] After training is complete, the test set is used to evaluate the performance of the model, and the detection precision (Precision), recall (Recall), F1-score, etc. are calculated. According to the evaluation results, the model can be optimized by adjusting the network structure, increasing the training data or performing hyperparameter tuning. The calculation formula of the evaluation index is as follows:

[0045] Precision:

[0046] Recall:

[0047] F1-score:

[0048] F1-score is a metric used to measure the performance of a classification model. It is the harmonic mean of precision and recall, combining the strengths of both, with F1-score values ranging between 0 and 1, with values closer to 1 indicating better model performance. It seeks a balance between precision and recall, making it particularly suitable for class-imbalanced datasets.

[0049] where TP, FP, and FN represent true positives, false positives, and false negatives, respectively.

[0050] Step 3: Based on the dataset, crop images containing defects and classify them to form a defect image dataset. Use the trained localization model to locate defect regions in the sample dataset and crop these regions to form new images. Each cropped image should contain only one defect as much as possible to improve the accuracy of subsequent classification. During the cropping process, the minimum bounding rectangle method can be used to accurately crop the defect region and improve the quality of the cropped images. Alternatively, use the cv2.boundingRect function in OpenCV for cropping.

[0051] Classify the cropped images according to different types of defects to form a multi-class defect image dataset. This classification can be based on features such as shape, size, and color of the defects, using manual or semi-automatic methods. Feature engineering-based methods can be introduced to extract texture features (such as LBP, GLCM, etc.) and shape features (such as Hu moments, Fourier descriptors, etc.) of defect images to improve classification accuracy.

[0052] Store the classified defect images according to their categories to form a defect image dataset with clear structure and explicit labels, ensuring that the sample size of each category is sufficient and as balanced as possible to facilitate subsequent model training. During storage, use a distributed file system (such as Hadoop HDFS) to improve data reliability and scalability.

[0053] Step 4: Model and train the image classifier with the defect map dataset and connect it with the positioning model; select a suitable image classification algorithm, such as a convolutional neural network (CNN), common network structures include VGG, ResNet, Inception, etc. The image classifier contains multiple convolutional layers, pooling layers, and fully connected layers, as well as Dropout layers to prevent overfitting. Cross-validation method is used to optimize network structure and parameters. Migration learning can be based on pre-trained models to improve training efficiency and model performance. When designing the classification network, attention mechanisms (such as SE-Block, CBAM, etc.) can be introduced to enhance the model's focus on key features and improve classification performance.

[0054] Preprocessing of the defect map dataset is also required before training, including image normalization, size adjustment, and data augmentation. Then build the image classifier and train it using the preprocessed defect map dataset. Set appropriate hyperparameters and select the best model through cross-validation. The specific method of cross-validation can use K-fold cross-validation, commonly used K values are 5 or 10, through multiple training and validation, improve the stability of the model.

[0055] Use the validation set and test set to evaluate the performance of the image classifier, calculate the classification accuracy, confusion matrix and other indicators. According to the evaluation results, optimize the model, adjust the network structure or perform hyperparameter tuning. During the evaluation process, AUC-ROC curve can be used to evaluate the classification ability of the model, the higher the AUC value, the better the model performance.

[0056] Step 5: Connect the trained positioning model and image classifier to form a complete defect detection and classification system. Input an image and output the bbox of the defect and the large category name. According to the factory attributes and process attributes, further specific classification of large category defects is carried out. Through the image classifier and positioning model, the defects are mapped, and computer vision post-processing is required before mapping to specific defect names. The factory attributes and process attributes include but are not limited to process parameters, machine number, product model, etc. The computer vision post-processing includes image enhancement, feature extraction and rule-based classification algorithm.

[0057] Optionally, a computer vision post-processing section post-processes the results output by the image classifier, including image enhancement (such as contrast adjustment, edge enhancement, etc.) and feature extraction (such as edge detection, texture analysis, etc.). Further classification is performed on the extracted features using rule-based classification algorithms to improve the accuracy of defect naming. The detected defects are mapped to specific defect names based on factory attributes (such as process parameters, machine number, product model, etc.) and process attributes. A mapping rule library can be constructed or a database can be used to store and query related information. The positioning model and image classifier are integrated into the production line quality monitoring system to realize automated defect detection and classification. Ensure the real-time and stability of the system, and timely discover and report the defects in the production process. In the actual production environment, the system is verified to ensure its accuracy and reliability. Regularly maintain and upgrade the system, including retraining of models, updating of rule libraries, etc., to adapt to changes in the production line and new types of defects.

[0058] The embodiment also discloses a panel production line defect detection general model modeling detection system, which comprises:

[0059] 1. Data collection and annotation module

[0060] Function: This module is responsible for real-time acquisition of high-resolution defect images from the production line and extraction of images from the historical database. It is also responsible for detailed defect annotation of the images.

[0061] Workflow:

[0062] Use a high-resolution camera with multispectral and HDR imaging technology to capture images in real time.

[0063] Classify the defects generated at each process stage (coating, exposure, etching, development, etc.).

[0064] Collect metadata of the images (acquisition time, equipment number, product model, etc.).

[0065] Use image annotation tools to annotate the position, shape, and type of defects and store them in XML or JSON format.

[0066] Implement a double annotation mechanism to ensure the quality of annotation and form a complete sample data set.

[0067] 2. Positioning model training module

[0068] Function: This module uses target detection algorithms to model and train the sample data set to create a model that can accurately locate defects.

[0069] Workflow:

[0070] Select appropriate target detection algorithm (YOLO, SSD, Faster R-CNN) or create a hybrid network structure.

[0071] Preprocess the sample dataset (normalization, scaling, data augmentation).

[0072] Train the model in a multi-GPU environment using cross-entropy loss and IoU loss functions.

[0073] Apply dynamic learning rate adjustment strategies (such as cosine annealing scheduling).

[0074] Evaluate model performance using validation and test sets and optimize.

[0075] 3. Defect Classification Module

[0076] Function: This module is responsible for cropping image regions containing defects and classifying these regions to form a defect map dataset.

[0077] Workflow:

[0078] Use the localization model to locate defect regions in the sample dataset and crop them to form new images.

[0079] Classify the cropped images based on features such as shape, size, and color of the defects.

[0080] Store the classified defect images, ensuring sufficient and balanced sample size.

[0081] 4. Defect Detection Module

[0082] Function: This module combines the localization model and image classifier to form a complete defect detection and classification system.

[0083] Workflow:

[0084] Map defects based on factory attributes and process attributes using the image classifier and localization model.

[0085] Perform computer vision post-processing, including image enhancement and feature extraction.

[0086] Use rule-based classification algorithms to classify features and improve the accuracy of defect naming.

[0087] Construct a mapping rule library or use a database to store and query relevant information.

[0088] Integrate into the production line's quality monitoring system to achieve automated defect detection and classification.

[0089] Through the detailed description of the above modules, we can ensure the efficiency and adaptability of the panel production line defect detection general model modeling detection system, and also can timely discover and report the defects that occur in the production process, to ensure product quality. Regular system maintenance and upgrade will further improve the accuracy and reliability of the system to adapt to the changes of the production line and the new types of defects. Such system design not only improves the production efficiency, but also reduces the cost and error rate of manual detection.

[0090] In addition, the present disclosure also provides a computer storage medium, which stores a computer program. When the computer program is run by a processor, the method in the foregoing embodiments is implemented.

[0091] Optionally, the computer readable storage medium can be FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM memory, etc. It can also be various devices including one or any combination of the above memories. The computer can be various computing devices including smart terminals and servers.

[0092] Optionally, the executable instructions can be in the form of programs, software, software modules, scripts or codes, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be deployed in any form, including as standalone programs or as modules, components, subroutines or other units suitable for use in computing environments.

[0093] As an example, the executable instructions can but not necessarily correspond to files in a file system, can be stored in part of a file that stores other programs or data, for example, in one or more scripts stored in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple cooperative files (for example, files storing one or more modules, subroutines or code portions).

[0094] As an example, the executable instructions can be deployed to execute on one computing device, or on multiple computing devices located at one site, or on multiple computing devices distributed in multiple sites and interconnected through a communication network.

[0095] It should be noted that, in this document, the terms "comprise", "comprising", or any other variant thereof are intended to cover a non-exclusive inclusion, such that processes, methods, articles, or systems that comprise a list of elements do not include only those elements recited, but can also include other elements not expressly listed or inherent to such processes, methods, articles, or systems. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or system that includes the element.

[0096] The above-mentioned sequence numbers of the embodiments of the present disclosure are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0097] Through the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, of course, they can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, an optical disk), and includes a plurality of instructions for causing a multimedia terminal device (which can be a mobile phone, a computer, a television receiver, or a network device, etc.) to execute the methods described in various embodiments of the present disclosure.

[0098] The above is only the preferred embodiment of the present disclosure, and does not limit the patent scope of the present disclosure, and any equivalent structure or equivalent process transformation made by using the content of the present disclosure specification and drawings, or directly or indirectly applied to other related technical fields, are also included in the patent protection scope of the present disclosure. Industrial applicability

[0099] The embodiment of the present disclosure provides a panel production line defect detection general model modeling detection method, system, device and storage medium, which can ensure the efficiency and adaptability of the panel production line defect detection general model modeling detection system, and also can timely find and report the defects in the production process, and ensure the product quality. Regular system maintenance and upgrading will further improve the accuracy and reliability of the system to adapt to the changes of the production line and new defect types. Thus, the production efficiency is improved, and the cost and error rate of manual detection are reduced.

Claims

1. A panel production line defect detection general model modeling detection method, characterized in that, The method comprises the steps of: Step 1, collecting panel production line sample data and defect labeling to obtain a sample data set; Step 2, modeling and training a positioning model based on the data set; Step 3, cropping images containing defects based on the data set and classifying to form a defect image data set; Step 4, modeling and training an image classifier based on the defect image data set, and connecting the image classifier with the positioning model; Step 5, mapping defects to specific defect names according to factory attributes and process attributes through the image classifier and the positioning model: The trained positioning model and image classifier are connected to form a complete defect detection and classification system, which inputs an image and outputs the bbox of the defect and the general category name. According to the factory attributes and process attributes, the general category defects are further specifically distinguished. The defects are mapped through the image classifier and the positioning model. Computer vision post-processing is required before mapping. The specific defect name is mapped. The factory attributes and process attributes include but are not limited to process parameters, machine number, and product model. The computer vision post-processing includes image enhancement, feature extraction, and rule-based classification algorithm. The extracted features are further classified using the rule-based classification algorithm.

2. The panel line defect detection general model modeling detection method of claim 1, wherein, The panel production line sample data includes panel defect images of multiple different processes and production lines.

3. The panel line defect detection general model modeling detection method of claim 1, wherein, In step 2, the data set is modeled using a target detection algorithm.

4. The panel line defect detection general model modeling detection method according to claim 3, wherein, The target detection algorithm is a YOLO, SSD, or Faster R-CNN model.

5. The panel line defect detection general model modeling detection method of claim 1, wherein, In step 5, the factory attributes and process attributes include but are not limited to process parameters, machine number, and product model.

6. A panel line defect detection general model modeling detection system, characterized in that, It comprises: a data collection and labeling module configured to collect panel production line sample data and perform defect labeling to obtain a sample data set; a positioning model training module configured to train a positioning model; a defect classification module configured to crop images containing defects and classify to form a defect image data set, and model and train an image classifier based on the defect image data set; a defect detection module configured to connect the positioning model and the image classifier to map defects to specific defect names, i.e. to connect the trained positioning model and the image classifier to form a complete defect detection and classification system, which inputs an image and outputs the bbox of the defect and the general category name. According to the factory attributes and process attributes, the general category defects are further specifically distinguished. The defects are mapped through the image classifier and the positioning model. Computer vision post-processing is required before mapping. The specific defect name is mapped. The factory attributes and process attributes include but are not limited to process parameters, machine number, and product model. The computer vision post-processing includes image enhancement, feature extraction, and rule-based classification algorithm. The extracted features are further classified using the rule-based classification algorithm.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The processor executes a computer program to implement the panel production line defect detection general model modeling and detection method of any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the panel production line defect detection general model modeling detection method in any one of claims 1-5.

Images