TFT display defect intelligent identification system based on machine vision

Abstract

The invention provides a TFT display defect intelligent identification system based on machine vision, and relates to the field of display defect intelligent identification, and the system comprises an image data acquisition module which is used for obtaining image data of a TFT display; the image data preprocessing module is used for preprocessing the image data; the feature extraction module is used for carrying out feature extraction on the preprocessed image data by using a gray scale run length matrix so as to obtain the features of the image data; the defect identification module is used for performing identification by using a transfer learning algorithm according to the characteristics of the image data so as to obtain a defect identification result; the deep learning module is used for classifying defect recognition results by using a BP neural network and performing continuous training by using defect samples so as to obtain an optimized BP neural network model; and the early warning module is used for obtaining early warning information according to the defect identification module and the defect threshold value. The problem that only specific types of defects can be detected can be effectively solved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent recognition of display defects, and in particular to a machine vision-based intelligent recognition system for TFT display defects. Background Art

[0002] In the field of display manufacturing, especially in the manufacturing process of TFT (Thin Film Transistor) displays, defect detection is a key link to ensure product quality. Traditional defect detection technologies mainly include manual visual inspection and mechanical contact inspection. Manual visual inspection has a low sampling rate and low accuracy. It is often difficult to achieve a 100% sampling rate for manual inspection, and it is greatly affected by the subjective factors of the inspectors. The accuracy is difficult to guarantee. The manual inspection speed is slow and cannot meet the needs of high-speed automated production lines. Long-term visual inspection work can easily lead to visual fatigue of inspectors, and the consistency of inspection standards is poor. Different inspectors may have different judgment standards for the same defect, resulting in inconsistent inspection results. Manual inspection cannot record the inspection process and results, which is not conducive to subsequent product traceability and quality analysis. Mechanical contact inspection method, high-quality mechanical inspection equipment is often expensive, and the mechanical contact inspection method can usually only detect specific types of defects, and the inspection speed is relatively slow. Since it needs to be in direct contact with the product, the mechanical device may cause secondary damage to the product during the inspection process. Summary of the invention

[0003] The present invention provides a TFT display defect intelligent recognition system based on machine vision, which can effectively solve the problem that only specific types of defects can be detected.

[0004] On the one hand, the present invention provides a TFT display defect intelligent recognition system based on machine vision, comprising: An image data acquisition module, used to acquire image data of the TFT display; An image data preprocessing module, used for preprocessing the image data to obtain preprocessed image data; A feature extraction module is used to extract features from the preprocessed image data using a grayscale run-length matrix to obtain features of the image data; The defect recognition module is used to use the transfer learning algorithm to perform recognition based on the characteristics of the image data to obtain defect recognition results; Deep learning module, used to classify defect recognition results using BP neural network, and use defect samples for continuous training to obtain an optimized BP neural network model; The early warning module is used to obtain early warning information based on the defect identification module and the defect threshold.

[0005] Furthermore, the image data acquisition module includes: An image data acquisition module, used for acquiring image data of the TFT display; Grayscale conversion module, used to convert image data into grayscale image.

[0006] Furthermore, the image data preprocessing module includes: A denoising module, used for denoising the grayscale image using an average filtering algorithm to obtain a denoised grayscale image; The contrast enhancement module is used to enhance the contrast of the denoised grayscale image to obtain a contrast-enhanced grayscale image, that is, preprocessed image data.

[0007] Furthermore, the feature extraction module includes: The grayscale run-length matrix construction module is used to determine the grayscale levels of the image and the expected maximum run-length according to the preprocessed image data, and to construct the grayscale run-length matrix, that is, ,in, Represents the grayscale run matrix in grayscale i , Run length j and direction θ The count value on Represents the pixel value of the image at position (m,n); The grayscale run-length matrix module is used to extract features using the grayscale run-length matrix to obtain features of image data.

[0008] Further, the defect identification module comprises; Construct a transfer learning model module to use the features of image data Perform model training to obtain a transfer learning model, where: is the number of training samples, is the number of categories, It is a sample Belongs to category The real label, is the model prediction sample Belongs to category The probability of The recognition module is used to use the transfer learning model to perform defect recognition on the features of the image data to obtain defect recognition results.

[0009] Furthermore, the deep learning module includes: BP neural network construction module, used to determine the number of layers of the BP neural network, the number of neurons in each layer, the activation function, and the weight and bias of the BP neural network according to the requirements of defect identification; The forward propagation module is used to calculate the prediction results of the output layer layer by layer using the BP neural network according to the defect recognition results; A loss calculation module is used to calculate the error between the prediction result and the defect recognition result using the cross entropy loss function; The back propagation module is used to calculate the gradient of each weight and bias through the back propagation algorithm according to the error, and update the weights and biases of the neural network; The iterative module repeats the process of forward propagation, loss calculation, back propagation and weight update until the preset threshold is reached to obtain the optimized BP neural network model.

[0010] Furthermore, the early warning module includes: A real-time monitoring module is used to obtain image data of the TFT display in real time and perform defect recognition to obtain real-time defect recognition results; A comparison module is used to compare the real-time defect recognition result with the preset defect threshold to obtain a comparison result; The judgment module is used to trigger the warning information according to the comparison result. If the comparison result exceeds or reaches the warning triggering condition, the warning information will not be triggered if the comparison result does not reach the warning triggering condition.

[0011] The machine vision-based TFT display defect intelligent recognition system provided by the present invention can obtain image data of the TFT display with high precision through the image data acquisition module, providing a reliable basis for subsequent processing; the feature extraction module adopts the grayscale run matrix method to effectively extract key features in the image, providing a strong basis for defect recognition; the defect recognition module combines with the transfer learning algorithm to quickly adapt to different types of defect recognition tasks, thereby improving the efficiency and accuracy of recognition; the deep learning module uses the BP neural network to classify the defect recognition results, and continuously optimizes the model through continuous training, so that the system can adaptively learn and recognize new defect types, thereby enhancing the adaptability and flexibility of the system; the system can adapt to different production requirements and quality control standards by adjusting the defect threshold to meet diversified production needs; the early warning module generates early warning information in time according to the defect recognition results and the preset defect threshold, reminds the operator to deal with potential quality problems in time, effectively avoids the outflow of defective products, and improves the quality control level of the production line; the system can monitor the operation status of the production line in real time, timely discover and warn abnormal situations in the production process, and provide strong support for production management; the automated and intelligent defect recognition system reduces manual intervention and reduces labor costs, while improving production efficiency and product yield, creating greater economic benefits for the enterprise. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1 It is a schematic diagram of a TFT display defect intelligent identification system based on machine vision provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] like Figure 1 As shown, the TFT display defect intelligent recognition system based on machine vision provided by the embodiment of the present invention mainly includes: An image data acquisition module, used to acquire image data of the TFT display; An image data preprocessing module, used for preprocessing the image data to obtain preprocessed image data; A feature extraction module is used to extract features from the preprocessed image data using a grayscale run-length matrix to obtain features of the image data; The defect recognition module is used to use the transfer learning algorithm to perform recognition based on the characteristics of the image data to obtain defect recognition results; Deep learning module, used to classify defect recognition results using BP neural network, and use defect samples for continuous training to obtain an optimized BP neural network model; The early warning module is used to obtain early warning information based on the defect identification module and the defect threshold.

[0016] In the embodiment of the present invention, the image data acquisition module is responsible for acquiring the image data of the TFT display. Through the high-precision image acquisition device, it can capture the subtle changes on the display surface and provide rich information for subsequent processing; the authenticity and integrity of the image data are ensured, and a reliable basis is provided for subsequent preprocessing and feature extraction; high-quality image data helps to improve the accuracy and efficiency of defect recognition; the image data preprocessing module preprocesses the collected image data, including denoising, enhancement, correction and other operations to eliminate interference factors in the image and improve the image quality; the preprocessed image data is clearer and easier to process, which provides favorable conditions for subsequent feature extraction and defect recognition; the processability and reliability of the image data are significantly improved. The preprocessed image data can more accurately reflect the real state of the display, which helps to improve the accuracy of defect recognition. The feature extraction module uses algorithms such as grayscale run-length matrix to extract features from the preprocessed image data to obtain feature vectors that can characterize image defects. Feature extraction is a key step in defect recognition, which can convert image data into information that can be understood and processed by computers. The extracted feature vectors can effectively characterize the defect information of the image, providing strong support for subsequent defect recognition. Through feature extraction, the dimension of image data can be greatly reduced, and processing efficiency and accuracy can be improved. The defect recognition module uses transfer learning algorithms to identify defects based on the extracted image features.Transfer learning can use existing knowledge and experience to quickly adapt to new recognition tasks; the defect recognition module can identify various defects in the image, such as bright spots, dark spots, lines, etc., and give preliminary recognition results; it improves the accuracy and efficiency of defect recognition and can quickly identify defects in images; the application of transfer learning algorithms enables the system to continuously learn and adapt to new defect types, improving the system's generalization and adaptability; the deep learning module uses BP neural network to classify defect recognition results, and further uses defect samples for continuous training to optimize the BP neural network model; the deep learning module can learn deeper feature representations and improve the accuracy and robustness of defect recognition; through continuous training and optimization, the BP neural network model can more accurately classify defect recognition results and improve the overall performance of the system; the application of deep learning modules enables the system to handle more complex The complex and diverse defect types improve the recognition and generalization capabilities of the system; the early warning module generates early warning information based on the output results of the defect recognition module and the set defect threshold; when the identified defects exceed the set threshold, the early warning module will issue a warning signal; the early warning module can promptly detect and report potential defect problems, and provide timely maintenance instructions for maintenance personnel; improve the reliability and safety of the system, and can promptly detect and deal with potential defect problems; the issuance of early warning information provides maintenance personnel with clear maintenance goals and directions, which helps to quickly locate and solve problems; in summary, the TFT display defect intelligent recognition system based on machine vision realizes efficient and accurate recognition of TFT display defects through the collaborative work of various modules; it not only improves production efficiency and product quality, but also provides timely maintenance instructions for maintenance personnel, and has broad application prospects and practical value.

[0017] like Figure 1 As shown, the image data acquisition module includes: An image data acquisition module, used for acquiring image data of the TFT display; Grayscale conversion module, used to convert image data into grayscale image.

[0018] In the embodiment of the present invention, the image data acquisition module is the input end of the system, which is responsible for capturing the image data of the TFT display in real time or periodically through a high-resolution camera or other image acquisition device; it ensures that the system can obtain clear and complete display images, providing a reliable basis for subsequent processing and analysis; it ensures that the source of image data is stable and of high quality, providing strong support for subsequent image processing and defect identification; through real-time or periodic image acquisition, the system can promptly detect any changes or abnormalities on the display surface, providing an important basis for quality control and troubleshooting in the production process; the grayscale conversion module is responsible for converting the acquired color image data into a grayscale image; the grayscale image only contains brightness information, not color information, so it is simpler and more efficient to process; the conversion process can reduce the complexity of image data It reduces the complexity and computational complexity of subsequent processing while retaining sufficient defect information for subsequent identification. It improves the efficiency and speed of image processing because the amount of data in grayscale images is much smaller than that in color images, so they are processed more quickly. The grayscale converted image is easier to analyze and process because the brightness information is more direct and obvious, which is helpful for subsequent defect identification and classification. Through grayscale conversion, the system can focus more on processing brightness changes related to defects without being disturbed by color information, thereby improving the accuracy and reliability of defect identification. In summary, the image data acquisition module and the grayscale conversion module in the image data acquisition module jointly undertake the important task of acquiring and processing TFT display image data. Through collaborative work, it ensures that the system can acquire high-quality, easy-to-process image data, providing a solid foundation for subsequent defect identification and classification.

[0019] like Figure 1 As shown, the image data preprocessing module includes: A denoising module, used for denoising the grayscale image using an average filtering algorithm to obtain a denoised grayscale image; The contrast enhancement module is used to enhance the contrast of the denoised grayscale image to obtain a contrast-enhanced grayscale image, that is, preprocessed image data.

[0020] In the embodiment of the present invention, the core task of the denoising module is to use the average filtering algorithm to remove noise from the grayscale image. During the image acquisition process, due to the interference of various external factors (such as device jitter, light changes, etc.), noise is often mixed into the original image, and the noise will interfere with subsequent image analysis and defect recognition; the average filtering algorithm replaces the original value of each pixel in the image by calculating the average value of the neighboring pixels around the pixel, thereby achieving the purpose of smoothing the image and eliminating noise; the grayscale image after denoising has a smoother surface and the noise is effectively suppressed, which helps to improve the accuracy and stability of subsequent image processing; the reduction of noise makes the defect information in the image more prominent, providing the defect recognition module with clearer image data, thereby improving the accuracy and efficiency of defect recognition; the contrast enhancement module is designed to improve the grayscale image after denoising Image contrast, which is a key factor in image clarity and recognizability; adjust the grayscale value distribution of the image to make the bright areas in the image brighter and the dark areas darker, thereby enhancing the image contrast; the grayscale image after contrast enhancement has clearer details and significantly improves the contrast between defects and background, which makes defects easier to identify in the image; the improved image contrast helps the defect recognition module to more accurately capture and identify defect information in the image, thereby improving the defect recognition capability and accuracy of the entire system; in summary, the denoising module and the contrast enhancement module in the image data preprocessing module work together to effectively improve the quality of the original grayscale image, providing clearer and easier-to-process image data for subsequent defect recognition; it not only improves the accuracy and efficiency of defect recognition, but also lays a solid foundation for the stability and reliability of the entire intelligent recognition system.

[0021] like Figure 1 As shown, the feature extraction module includes: The grayscale run-length matrix construction module is used to determine the grayscale levels of the image and the expected maximum run-length according to the preprocessed image data, and to construct the grayscale run-length matrix, that is, ,in, Represents the grayscale run matrix in grayscale i , Run length j and direction θ The count value on Represents the pixel value of the image at position (m,n); The grayscale run-length matrix module is used to extract features using the grayscale run-length matrix to obtain features of image data.

[0022] In an embodiment of the present invention, a grayscale run matrix construction module is responsible for first determining the grayscale level of the image, that is, the number of different grayscale values ​​in the image, and the expected maximum run length, that is, the maximum number of pixels with the same continuous grayscale value in the image, based on the preprocessed image data; then, the grayscale run matrix is ​​constructed using the information, which is a special matrix structure that can efficiently represent the distribution and continuity of grayscale values ​​in the image; the grayscale run matrix not only considers the size of the grayscale value, but also considers the continuity and spatial distribution of the grayscale value, so it can more comprehensively describe the characteristics of the image; by constructing the grayscale run matrix, the system can convert complex image data into a structured feature representation, which provides great convenience for subsequent feature extraction and defect recognition; the construction process of the grayscale run matrix fully considers the spatial information of the image, so that the extracted features are more representative and discriminative, which helps to improve the accuracy of defect recognition; the grayscale run matrix module uses the constructed grayscale run matrix to extract features, and by analyzing the elements in the matrix, such as The length, number, and distribution of the run are used to extract features that can characterize image defects; these may include the shape, size, location, and abnormal changes in grayscale values ​​of the defects, which are crucial for subsequent defect identification; through the feature extraction of the grayscale run matrix module, the system can accurately capture the defect information in the image, providing a strong basis for defect identification; the extracted features are highly accurate and stable, and can maintain high recognition performance even under complex image backgrounds and noise interference; the feature extraction process is efficient and automated, which greatly improves the efficiency and accuracy of defect identification, and provides strong technical support for quality control and troubleshooting of TFT displays; in summary, the feature extraction module realizes the efficient and accurate extraction of features that are crucial to defect identification from the preprocessed image data by constructing the collaborative work of the grayscale run matrix module and the grayscale run matrix module; this not only improves the accuracy and efficiency of defect identification, but also lays a solid foundation for the stability and reliability of the entire intelligent recognition system.

[0023] like Figure 1 As shown, the defect identification module comprises; Construct a transfer learning model module to use the features of image data Perform model training to obtain a transfer learning model, where: is the number of training samples, is the number of categories, It is a sample Belongs to category The real label, is the model prediction sample Belongs to category The probability of The recognition module is used to use the transfer learning model to perform defect recognition on the features of the image data to obtain defect recognition results.

[0024] In an embodiment of the present invention, a transfer learning model module is constructed to utilize a large amount of existing data and knowledge in related fields, and this knowledge and experience are transferred to a new TFT display defect recognition task through transfer learning technology, thereby avoiding the large amount of data and computing resources required for training the model from scratch, and accelerating the development and deployment of the model; the transfer learning model can learn a more generalized feature representation, so that the model can still maintain a high recognition accuracy when facing new, unseen defect types or image changes, thereby enhancing the system's adaptability to different production environments and product changes; through transfer learning, the model can achieve better performance on less training data, significantly reducing the time and computing cost required for model training; the recognition module uses the transfer learning model to perform defect recognition on the pre-processed image data features, and can accurately identify various defects in the TFT display, such as bright spots, dark spots, line defects, etc. This improves the quality control level of the production line and reduces the outflow of defective products. Since the transfer learning model has learned a lot of relevant knowledge, the recognition module can quickly identify new image data in a shorter time, meeting the real-time requirements of high-speed automated production lines. At the same time, the efficient recognition process also improves the overall efficiency of the production line. The recognition module can be easily integrated into the existing production system and seamlessly connected with other modules (such as image data acquisition, preprocessing, feature extraction, etc.) to form a complete defect recognition process. As production needs change or new defect types emerge, the recognition module can adapt to new recognition tasks by updating the transfer learning model, maintaining the flexibility and scalability of the system.

[0025] like Figure 1 As shown, the deep learning module includes: BP neural network construction module, used to determine the number of layers of the BP neural network, the number of neurons in each layer, the activation function, and the weight and bias of the BP neural network according to the requirements of defect identification; The forward propagation module is used to calculate the prediction results of the output layer layer by layer using the BP neural network according to the defect recognition results; A loss calculation module is used to calculate the error between the prediction result and the defect recognition result using the cross entropy loss function; The back propagation module is used to calculate the gradient of each weight and bias through the back propagation algorithm according to the error, and update the weights and biases of the neural network; The iterative module repeats the process of forward propagation, loss calculation, back propagation and weight update until the preset threshold is reached to obtain the optimized BP neural network model.

[0026] In the embodiment of the present invention, the BP neural network construction module is responsible for designing and constructing the structure of the BP neural network according to the specific needs of defect recognition, including determining the number of network layers, the number of neurons in each layer, selecting a suitable activation function, and initializing the weights and biases of the network; reasonable network structure design is the key to ensuring that the neural network can effectively learn and accurately identify defects; through a carefully designed network structure, the BP neural network can process input data more efficiently and improve the accuracy of defect recognition; appropriate activation functions and weight initialization help avoid the problem of gradient disappearance or explosion during training and ensure stable learning of the network; the forward propagation module is responsible for transmitting the defect recognition results (i.e., input data) through the BP neural network The network performs forward propagation calculations, passes and processes data layer by layer, and finally obtains the prediction results of the output layer; forward propagation is the basic process of neural network prediction and classification; through forward propagation, the system can quickly obtain the prediction results of the input data, providing a basis for subsequent loss calculation and back propagation; efficient forward propagation calculations help to improve the processing speed of the entire system; the loss calculation module uses the cross entropy loss function to calculate the error between the prediction result and the actual defect recognition result; the cross entropy loss function is a common method to measure the difference between the predicted probability distribution and the actual probability distribution in the classification task; accurate loss calculation is the key to guiding the neural network to update the weights; through the cross entropy loss function, the system can quantify the predicted The error between the measured result and the actual result provides a clear optimization target for back propagation; reasonable loss calculation helps the neural network converge faster during the training process and improves the training efficiency; the back propagation module calculates the gradient of each weight and bias through the back propagation algorithm based on the error obtained by loss calculation, indicating how the network weight should be adjusted to reduce the error; back propagation is the core process of neural network weight update; through the back propagation algorithm, the system can accurately calculate the gradient of each weight and bias, providing precise guidance for weight update; effective back propagation helps the neural network continuously optimize its parameters during the training process and improve the accuracy of defect recognition; the iteration module is responsible for repeatedly executing forward propagation and loss calculation , back propagation and weight update process; the process will continue until the preset threshold is reached (such as the loss value is small enough or the number of iterations reaches the upper limit); through iterative training, the BP neural network can gradually learn the mapping relationship from input data to output results; through continuous iterative training, the BP neural network can gradually optimize its weights and biases, improve the accuracy and generalization ability of defect recognition; iterative training enables the neural network to adapt to different defect types and complex backgrounds, and improve the robustness and practicality of the system; in summary, the deep learning module realizes efficient and accurate recognition of TFT display defects through the collaborative work of sub-modules such as BP neural network construction, forward propagation, loss calculation, back propagation and iterative training;It not only improves the accuracy of defect identification, but also enhances the adaptability and robustness of the system, providing strong technical support for quality control and troubleshooting of TFT displays.

[0027] like Figure 1 As shown, the early warning module includes: A real-time monitoring module is used to obtain image data of the TFT display in real time and perform defect recognition to obtain real-time defect recognition results; A comparison module is used to compare the real-time defect recognition result with the preset defect threshold to obtain a comparison result; The judgment module is used to trigger the warning information according to the comparison result. If the comparison result exceeds or reaches the warning triggering condition, the warning information will not be triggered if the comparison result does not reach the warning triggering condition.

[0028] In the embodiment of the present invention, the real-time monitoring module is responsible for continuously acquiring the image data of the TFT display, and using the aforementioned image data acquisition, preprocessing, feature extraction and defect recognition technologies to perform real-time analysis on the image to obtain real-time defect recognition results; ensuring that the system can promptly detect any abnormalities or defects on the display, and provide accurate data support for subsequent early warnings; through real-time monitoring, the system can quickly respond to any changes on the display, greatly improving the timeliness and accuracy of defect detection; real-time data acquisition and processing provide a reliable foundation for the early warning system, making the early warning information more accurate and reliable; the comparison module is responsible for comparing the defect recognition results obtained by the real-time monitoring module with the preset defect threshold; the threshold is usually set based on historical data, industry standards or user needs, and represents an acceptable defect range; through comparison Yes, the system can determine whether the current defect situation exceeds the acceptable range, providing a basis for subsequent early warning judgment; the existence of the comparison module enables the early warning system to trigger the early warning according to clear judgment criteria, avoiding false alarms or missed alarms; by comparing with the preset threshold, the system can more accurately locate potential problems, improving the accuracy and pertinence of the early warning; the judgment module decides whether to trigger the early warning information based on the results of the comparison module; when the comparison results show that the defects exceed or reach the conditions for triggering the early warning, the judgment module will immediately trigger the early warning; otherwise, the early warning will not be triggered; ensuring the timely and accurate issuance of early warning information, providing maintenance personnel with clear maintenance instructions; through the intelligent decision-making of the judgment module, the early warning system can issue early warning information at the most appropriate time, avoiding unnecessary interference and ensuring the timely discovery and handling of problems; Accurate warning information helps maintenance personnel quickly locate and solve problems, improving maintenance efficiency and display availability. In summary, the warning module realizes real-time monitoring and early warning of TFT display defects through the collaborative work of three sub-modules: real-time monitoring, comparison and judgment. It not only improves the timeliness and accuracy of defect detection, but also provides maintenance personnel with clear maintenance instructions, providing strong support for quality control and troubleshooting of displays.

[0029] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A TFT display defect intelligent recognition system based on machine vision, characterized in that: include: An image data acquisition module, used to acquire image data of the TFT display; An image data preprocessing module, used for preprocessing the image data to obtain preprocessed image data; A feature extraction module is used to determine the grayscale level and the expected maximum run length of the image according to the preprocessed image data, construct a grayscale run matrix, and use the grayscale run matrix to perform feature extraction to obtain the features of the image data; The defect recognition module is used to use the characteristics of the image data Model training is performed to obtain a transfer learning model, and the transfer learning model is used to identify defects based on the features of the image data to obtain defect identification results, where: is the number of training samples, is the number of categories, It is a sample Belongs to category The real label, is the model prediction sample Belongs to category The probability of Deep learning module, used to classify defect recognition results using BP neural network, and use Continuous training is performed to obtain the optimized BP neural network model, in which: represents the set of optimized weights and biases, is the current value of the set of weights and biases, L is the loss function, is the prediction result, y is the difference between the true labels, is the input feature matrix, The BP neural network is based on the input features and current parameters The predicted output results; The early warning module is used to obtain early warning information based on the defect identification module and the defect threshold.

2. The TFT display defect intelligent identification system based on machine vision according to claim 1 is characterized in that: The image data acquisition module comprises: An image data acquisition module, used for acquiring image data of the TFT display; Grayscale conversion module, used to convert image data into grayscale image.

3. The TFT display defect intelligent identification system based on machine vision according to claim 1 is characterized in that: The image data preprocessing module comprises: A denoising module, used for denoising the grayscale image using an average filtering algorithm to obtain a denoised grayscale image; The contrast enhancement module is used to enhance the contrast of the denoised grayscale image to obtain a contrast-enhanced grayscale image, that is, preprocessed image data.

4. The TFT display defect intelligent identification system based on machine vision according to claim 1 is characterized in that: The feature extraction module comprises: The grayscale run-length matrix construction module is used to determine the grayscale levels of the image and the expected maximum run-length according to the preprocessed image data, and to construct the grayscale run-length matrix, that is, ,in, Represents the grayscale run matrix in grayscale i , Run length j and direction θ The count value on Represents the pixel value of the image at position (m,n); The grayscale run-length matrix module is used to extract features using the grayscale run-length matrix to obtain features of image data.

5. The TFT display defect intelligent identification system based on machine vision according to claim 1 is characterized in that: The defect identification module comprises; Construct a transfer learning model module to use the features of image data Perform model training to obtain a transfer learning model, where: is the number of training samples, is the number of categories, It is a sample Belongs to category The real label, is the model prediction sample Belongs to category probability; The recognition module is used to use the transfer learning model to perform defect recognition on the features of the image data to obtain defect recognition results.

6. The TFT display defect intelligent identification system based on machine vision according to claim 1 is characterized in that: The deep learning module includes: BP neural network construction module, used to determine the number of layers of the BP neural network, the number of neurons in each layer, the activation function, and the weight and bias of the BP neural network according to the requirements of defect identification; The forward propagation module is used to calculate the prediction results of the output layer layer by layer using the BP neural network according to the defect recognition results; A loss calculation module is used to calculate the error between the prediction result and the defect recognition result using the cross entropy loss function; The back propagation module is used to calculate the gradient of each weight and bias through the back propagation algorithm according to the error, and update the weights and biases of the neural network; The iterative module repeats the process of forward propagation, loss calculation, back propagation and weight update until the preset threshold is reached to obtain the optimized BP neural network model.

7. The TFT display defect intelligent identification system based on machine vision according to claim 1 is characterized in that: The early warning module comprises: A real-time monitoring module is used to obtain image data of the TFT display in real time and perform defect recognition to obtain real-time defect recognition results; A comparison module is used to compare the real-time defect recognition result with the preset defect threshold to obtain a comparison result; The judgment module is used to trigger the warning information according to the comparison result. If the comparison result exceeds or reaches the warning triggering condition, the warning information will not be triggered if the comparison result does not reach the warning triggering condition.

Images