Panel defect detection method and system based on multi-model fusion

Abstract

The invention provides a panel defect detection method and system based on multi-model fusion, and relates to the technical field of panel defect detection, and the method comprises the steps: carrying out the first-stage defect detection of a panel sample image through employing a mode of combining edge detection and isolated forest determination, so as to obtain the defect of a screen display abnormality type; performing secondary defect detection on the panel sample image by adopting an image classification model to obtain defects of a screen scratch category; performing three-stage defect detection on the panel sample image by adopting the target detection model to obtain defects of screen dead pixels and bad line categories; and performing result fusion on different types of defects to obtain a panel defect detection result. According to the invention, a mode of combining an image processing technology, a machine learning technology and a neural network technology is adopted, and the problems that the defect of panel badness cannot be fully detected and missing detection is easy to occur due to the fact that an existing single algorithm or model based on a machine vision technology is difficult to cover multiple types of bad mobile phone panels are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of panel defect detection, and more specifically, to a panel defect detection method and system based on multi-model fusion. Background Art

[0002] In the production process of smart phones, the quality control of mobile phone panels is a crucial link. Mobile phone panels not only need to have good touch sensitivity and display effects, but also need to ensure that the appearance is free of defects to meet consumers' high standards for product performance and aesthetics. However, traditional manual inspection is inefficient and easily affected by subjective judgment, making it difficult to ensure the consistency and accuracy of inspection results.

[0003] With the rapid development of machine vision technology, more and more automated inspection devices have been introduced into the production line to improve the efficiency and accuracy of inspection. Machine vision technology simulates the visual function of the human eye and uses technologies such as optical lighting, imaging systems, and visual information processing to achieve defect detection of the surface defects of mobile phone panels. However, existing single algorithms or models based on machine vision technology are difficult to cover various categories of mobile phone panel defects, resulting in insufficient detection of panel defects and prone to missed detection problems. Summary of the Invention

[0004] The present invention provides a panel defect detection method and system based on multi-model fusion, which solves the problem that existing single algorithms or models based on machine vision technology are difficult to cover various categories of mobile phone panel defects, resulting in insufficient detection of panel defects and prone to missed detection problems.

[0005] In a first aspect, an embodiment of the present invention provides a panel defect detection method based on multi-model fusion, and the method includes the following processes:

[0006] Perform image smoothing processing, histogram equalization processing, and contrast enhancement processing on the original panel image to obtain a panel sample image;

[0007] Perform primary defect detection on the panel sample image by combining edge detection and isolation forest determination to obtain defects in the category of abnormal screen display;

[0008] Perform secondary defect detection on the panel sample image using an image classification model to obtain defects in the category of screen scratches;

[0009] Perform tertiary defect detection on the panel sample image using an object detection model to obtain defects in the categories of screen dead pixels and bad lines;

[0010] Fuse the results of defects in the screen display anomaly category, defects in the screen scratch category, and defects in the screen dead pixel and defective line category to obtain the panel defect detection result.

[0011] In the above embodiments, the present invention first preprocesses the panel image using traditional image processing techniques, and then combines the contour and size features corresponding to the defects in the screen display anomaly category, and determines whether there are screen display anomalies by combining edge detection and isolation forest determination; then combines the features corresponding to the defects in the screen scratch category, where the defect type and location are difficult to define, and uses an image classification model to determine whether there are defects in the screen scratch category. Then, combining the features corresponding to the defects in the screen dead pixel and defective line category, where the defect morphology is relatively unified, uses object detection to determine whether there are defects in the screen dead pixel and defective line category. Finally, fuse the defect detection results of different categories, and optimize the final determination result according to the definition rules of defective defects and the priority ranking of defects in different business scenarios.

[0012] As some alternative embodiments of the present application, the process of performing primary defect detection on the panel sample image by combining edge detection and isolation forest determination is as follows:

[0013] Based on the edge detection algorithm, perform edge contour detection on the panel sample image to obtain the maximum circumscribed rectangle of the panel;

[0014] Take the maximum circumscribed rectangle of the panel as the region of interest, and based on the isolation forest algorithm, determine whether the size of the region of interest is abnormal. If it is abnormal, it is determined that the panel has defects in the screen display anomaly category; otherwise, it is determined that the panel does not have defects in the screen display anomaly category.

[0015] In the above embodiments, the present invention can quickly and accurately determine whether there are defects in the screen display anomaly category by combining edge detection and the isolation forest algorithm.

[0016] As some alternative embodiments of the present application, the process of performing secondary defect detection on the panel sample image using an image classification model is as follows:

[0017] Input the panel sample image into the image classification model, and perform defect classification detection and normalization processing on the panel sample image through the image classification model to output the defect category probability of the screen scratch category;

[0018] Based on the threshold verification method, determine whether the defect category probability of the screen scratch category meets the verification standard. If it meets the verification standard, it is determined that the panel has defects in the screen scratch category; otherwise, it is determined that the panel does not have defects in the screen scratch category.

[0019] In the above embodiments, the present invention uses an image classification model to detect defects whose defect types and positions are difficult to define, that is, the method of whole-image classification is more accurate and robust.

[0020] As some alternative embodiments of the present application, the structure of the image classification model includes a shallow convolutional neural network and a normalization activation function.

[0021] As some alternative embodiments of the present application, the process of using an object detection model to perform three-level defect detection on a panel sample image is as follows:

[0022] Input the panel sample image into the object detection model, and perform feature localization and classification on the panel sample image through the object detection model to obtain object candidate boxes;

[0023] Perform non-maximum suppression processing on the object candidate boxes to finally obtain object candidate boxes with defects such as screen dead pixels and defective line categories.

[0024] In the above embodiments, the present invention uses an object detection model to detect defects with relatively uniform defect morphologies, and the method of object detection is more accurate and has better generalization ability.

[0025] As some alternative embodiments of the present application, the structure of the object detection model includes a deep convolutional neural network, a region proposal network, and a non-maximum suppression function.

[0026] As some alternative embodiments of the present application, the process of fusing the results of defects in the screen display anomaly category, the screen scratch category, and the screen dead pixel and defective line category is as follows:

[0027] Sort different category defects of the panel sample image according to the bad priority sorting rule;

[0028] Combine the confidence level to screen the sorted different category defects to obtain the final panel bad detection result.

[0029] In the above embodiments, the present invention fuses the defect detection results, and optimizes the final determination result according to the definition rules of bad defects and the bad priority sorting in different business scenarios.

[0030] In a second aspect, the present invention provides a panel bad detection system based on multi-model fusion, and the system includes:

[0031] An image preprocessing unit, which is used to perform image smoothing processing, histogram equalization processing, and contrast enhancement processing on the original panel image to obtain a panel sample image;

[0032] A display anomaly detection unit, which uses a combination of edge detection and isolation forest determination to perform primary defect detection on a panel sample image to obtain defects of the screen display anomaly category;

[0033] A screen scratch detection unit, which uses an image classification model to perform secondary defect detection on a panel sample image to obtain defects of the screen scratch category;

[0034] A dead pixel and defective line detection unit, which uses an object detection model to perform tertiary defect detection on a panel sample image to obtain defects of the screen dead pixel and defective line categories;

[0035] A defect result fusion unit, which is used to fuse the results of defects of the screen display anomaly category, defects of the screen scratch category, and defects of the screen dead pixel and defective line categories to obtain the panel defect detection result.

[0036] In a third aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the method for detecting panel defects based on multi-model fusion.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the method for detecting panel defects based on multi-model fusion.

[0038] The beneficial effects of the present invention are as follows: The present invention combines traditional image processing technology, machine learning technology, and neural network technology, and sets different defect detection models according to the characteristics of different panel defects. Through different models, various types of defect detections can be carried out quickly and accurately, and the defects of panel defects can be fully detected, and it is not easy to have missed detections and over-detections. Description of the Drawings

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can also be obtained based on these drawings.

[0040] Figure 1 It is a schematic diagram of the structure of a computer device in the hardware operating environment of the embodiments of the present invention;

[0041] Figure 2It is a flowchart of the panel defect detection method according to an embodiment of the present invention;

[0042] Figure 3 It is a structural block diagram of the panel defect detection system according to an embodiment of the present invention. Detailed implementation manners

[0043] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0044] In order to solve the problem that it is difficult for a single algorithm or model based on machine vision technology to cover various categories of defects in mobile phone panels, resulting in insufficient detection of panel defects and prone to missed detections. The present application provides a panel defect detection method and system based on multi-model fusion. Before introducing the specific technical solutions of the present application, the hardware operating environment involved in the embodiments of the present application is introduced first.

[0045] Please refer to Figure 1 , Figure 1 It is a schematic structural diagram of a computer device for the hardware operating environment involved in the embodiments of the present application.

[0046] As Figure 1 shown, the computer device may include: a processor, such as a central processing unit (CPU), a communication bus, a user interface, a network interface, and a memory. Among them, the communication bus is used to realize the connection and communication between these components. The user interface may include a display screen (Display) and an input unit such as a keyboard (Keyboard). Optionally, the user interface may further include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory may be a high-speed random access memory (Random Access Memory, RAM) or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. Optionally, the memory may also be a storage device independent of the aforementioned processor.

[0047] Those skilled in the art can understand that Figure 1 the structure shown in

[0048] does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine some components, or arrange different components. Figure 1 As

[0049] shown, the memory as a storage medium may include an operating system, a network communication module, a user interface module, and an electronic program module. Figure 1In the computer device shown, the network interface is mainly used for data communication with a network server; the user interface is mainly used for data interaction with a user; the processor and memory in the computer device of the present application can be set in the computer device, and the computer device calls the computer program product stored in the electronic program module through the processor and executes the panel defect detection method based on multi-model fusion provided in the embodiments of the present application.

[0050] Based on the hardware environment of the foregoing embodiments, an embodiment of the present application provides a panel defect detection method based on multi-model fusion. Please refer to Figure 2 , Figure 2 which is the flowchart of the panel defect detection method, and the method process is as follows:

[0051] (1) Perform image smoothing processing, histogram equalization processing, and contrast enhancement processing on the original panel image to obtain a panel sample image.

[0052] Specifically, the image smoothing processing includes but is not limited to mean filtering, Gaussian filtering, median filtering, etc. By performing image smoothing processing on the original panel image, the noise in the image can be reduced, making the original panel image look smoother.

[0053] Specifically, the calculation formula for the histogram equalization processing is as follows:

[0054]

[0055] where s k represents the value of the k-th gray level after histogram equalization processing, n i represents the number of pixels with gray level i in the histogram of the original panel image, N represents the total number of pixels in the histogram of the original panel image, and L represents the maximum value of the gray value after histogram equalization processing (for example, the maximum value of the gray value is 256).

[0056] Specifically, the calculation formula for the contrast enhancement processing is as follows:

[0057] g(x,y) = c * log(1 + f(x,y))

[0058] where g(x, y) represents the pixel value of the image at position (x, y) after contrast enhancement transformation, (x, y) represents the pixel value of the original panel image at position (x, y), c is a constant used to adjust the degree of contrast enhancement, it is a positive scaling factor that can control the intensity of the logarithmic transformation. A larger c value will increase the contrast, while a smaller c value will reduce the contrast.

[0059] In the embodiments of the present invention, image smoothing processing, histogram equalization processing, and contrast enhancement processing are performed on the original panel image using image processing techniques, which can facilitate subsequent defect detection processing, remove image noise, enhance the global contrast of the image, and local intensity.

[0060] (2) The panel sample image is subjected to primary defect detection by combining edge detection and the determination of the isolated forest of machine learning to obtain defects of the screen display abnormal category.

[0061] Specifically, in the embodiments of the present invention, the panel sample image can also be initially subjected to defect detection by combining edge detection and size threshold determination to obtain defects of the screen display abnormal category. That is, first, the maximum circumscribed rectangle of the panel is obtained through edge detection, and then it is judged whether the size of the maximum circumscribed rectangle is within the threshold determination range by the method of size threshold determination. If it is not within the threshold determination range, it is determined that the panel has defects of the screen display abnormal category; otherwise, it is determined that the panel does not have defects of the screen display abnormal category.

[0062] In the embodiments of the present invention, the process of performing primary defect detection on the panel sample image by combining edge detection and the determination of the isolated forest is as follows:

[0063] (2.1) Based on the edge detection algorithm, edge contour detection is performed on the panel sample image to obtain the maximum circumscribed rectangle of the panel.

[0064] In the embodiments of the present invention, first, the panel sample image is subjected to convolution operation using a convolution kernel. Through the convolution operation, the gradient value of each pixel point in the X direction and the gradient value in the Y direction in the panel sample image can be obtained; then, by calculating the gradient value in the X direction and the gradient value in the Y direction, the gradient amplitude and the gradient direction can be obtained.

[0065] Specifically, the calculation formulas for the gradient amplitude and the gradient direction are as follows:

[0066]

[0067] Among them, G x represents the gradient value in the x direction, G y represents the gradient value in the y direction, G represents the gradient amplitude, and θ represents the gradient direction.

[0068] Finally, based on the gradient amplitude and the gradient direction, non-maximum suppression and double-threshold detection are performed on the panel sample image, and the edges are connected through hysteresis thresholding to obtain the maximum circumscribed rectangle of the panel.

[0069] (2.2) Take the maximum circumscribed rectangle of the panel as the region of interest, and determine whether the size of the region of interest is abnormal based on the isolation forest algorithm. If there is an abnormality, it is determined that the panel has a defect in the screen display abnormality category; otherwise, it is determined that the panel does not have a defect in the screen display abnormality category.

[0070] In the embodiment of the present invention, first, collect the size data of the region of interest as sample data, and train the isolation forest algorithm model based on the sample data.

[0071] Specifically, the calculation formula of the isolation forest algorithm model is as follows:

[0072]

[0073] Among them, h(x) represents the path length of the data point x in a certain isolation tree, E(h(x)) represents the expected path length in all isolation trees, and c(n) is the average search path length of the binary sorting tree (BST) when the sample size is n, which is used to normalize the expected path length of the sample.

[0074] Specifically, the calculation formula of c(n) is as follows:

[0075]

[0076] Among them, n represents the number of samples used to generate each isolation tree, H (n-1) = 1 + 1 / 2 + … + 1 / n.

[0077] Then, use the trained isolation forest algorithm model to predict the size data of the new region of interest, and determine whether each data point is an abnormal point to obtain the prediction result; the prediction result will return an array.

[0078] Finally, analyze the abnormal characteristics and distribution of the data points according to the prediction result, and the proportion of data point abnormalities can be calculated. Based on the proportion of data point abnormalities, determine whether the size of the region of interest is abnormal. Among them, the proportion of data abnormal points can be set according to the actual situation, and the embodiment of the present invention does not limit this.

[0079] (3) Use an image classification model to perform secondary defect detection on the panel sample image to obtain defects in the screen scratch category.

[0080] Specifically, the structure of the image classification model includes a shallow (3 - layer) convolutional neural network (CNN) and a normalization activation function.

[0081] In the embodiment of the present invention, the process of using an image classification model to perform secondary defect detection on the panel sample image is as follows:

[0082] (3.1) Input the panel sample image into the image classification model. Use the shallow convolutional neural network of the image classification model to perform defect classification detection on the panel sample image, and perform normalization processing through the normalization activation function to output the defect category probability of the screen scratch category.

[0083] (3.2) Determine whether the defect category probability of the screen scratch category meets the verification standard based on the threshold verification method. If it meets the verification standard, it is determined that the panel has a defect of the screen scratch category; otherwise, it is determined that the panel does not have a defect of the screen scratch category. Among them, the threshold can be set according to the actual situation, and the embodiments of the present invention do not limit this.

[0084] (4) Use the object detection model to perform three-level defect detection on the panel sample image to obtain defects of the screen dead pixel and defective line categories.

[0085] Specifically, the structure of the object detection model includes a deep (101-layer) convolutional neural network, a region proposal network (RPN), and a non-maximum suppression function.

[0086] In the embodiments of the present invention, the process of using the object detection model to perform three-level defect detection on the panel sample image is as follows:

[0087] (4.1) Input the panel sample image into the object detection model. Use the deep convolutional neural network of the object detection model to extract features from the panel sample image, and extract object candidate boxes through the region proposal network, and achieve feature localization and classification through two output paths to obtain object candidate boxes.

[0088] (4.2) Perform non-maximum suppression processing on the object candidate boxes through the non-maximum suppression function to finally obtain the object candidate boxes with defects of the screen dead pixel and defective line categories.

[0089] Specifically, the structure of the object detection model includes a deep convolutional neural network, a region proposal network, and a non-maximum suppression function.

[0090] (5) Perform result fusion on the defects of the screen display anomaly category, the screen scratch category, and the screen dead pixel and defective line categories to obtain the panel defect detection result.

[0091] In the embodiments of the present invention, the process of performing result fusion on the defects of the screen display anomaly category, the screen scratch category, and the screen dead pixel and defective line categories is as follows:

[0092] (5.1) Sort different categories of defects in the panel sample image according to the sorting rules of defect priorities. For example, if the priority of the defect in the screen display anomaly category is higher than that of the defect in the screen scratch category, and the priority of the defect in the screen scratch category is higher than that of the defects in the screen dead pixel and defective line categories, then sort different categories of defects in the panel sample image according to the priority order.

[0093] (5.2) Screen out different categories of sorted defects in combination with the confidence level, and defective categories with low confidence levels can be removed to obtain the final panel defect detection result.

[0094] In summary, the present invention combines multiple algorithms such as image processing, machine learning, and convolutional neural network to detect various types of defects in mobile phone panels, solving the problem that it is difficult for a single algorithm or model based on machine vision technology to cover various types of defects in mobile phone panels, resulting in insufficient detection of panel defects and easy occurrence of missed detections.

[0095] In addition, in one embodiment, based on the same inventive concept as the foregoing embodiment, the embodiment of the present invention provides a panel defect detection system based on multi-model fusion. The system corresponds one-to-one with the method of Embodiment 1. Please refer to Figure 3 , Figure 3 which is the structural block diagram of the panel defect detection system. The system includes:

[0096] An image preprocessing unit, which is used to perform image smoothing processing, histogram equalization processing, and contrast enhancement processing on the original panel image to obtain a panel sample image;

[0097] A display anomaly detection unit, which uses a combination of edge detection and isolated forest determination to perform primary defect detection on the panel sample image to obtain defects in the screen display anomaly category;

[0098] A screen scratch detection unit, which uses an image classification model to perform secondary defect detection on the panel sample image to obtain defects in the screen scratch category;

[0099] A dead pixel and defective line detection unit, which uses an object detection model to perform tertiary defect detection on the panel sample image to obtain defects in the screen dead pixel and defective line categories;

[0100] A defect result fusion unit, which is used to fuse the results of defects in the screen display anomaly category, defects in the screen scratch category, and defects in the screen dead pixel and defective line categories to obtain the panel defect detection result.

[0101] It should be noted that in this embodiment, each unit in the panel defect detection system based on multi-model fusion corresponds one-to-one to each step in the panel defect detection method based on multi-model fusion in the foregoing embodiment. Therefore, the specific implementation manners and achieved technical effects of this embodiment can refer to the implementation manners of the foregoing panel defect detection method based on multi-model fusion, and will not be elaborated here.

[0102] In addition, in one embodiment, the present application further provides a computer device, which includes a processor, a memory, and a computer program stored in the memory. When the computer program is run by the processor, it implements the method in the foregoing embodiment.

[0103] In addition, in one embodiment, the present application further provides a computer storage medium, on which a computer program is stored. When the computer program is run by the processor, it implements the method in the foregoing embodiment.

[0104] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM; or it may be various devices including one or any combination of the above memories. The computer may be various computing devices including intelligent terminals and servers.

[0105] In some embodiments, the executable instructions may be in the form of a program, software, software module, script, or code, and may be written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including being deployed as an independent program or being deployed as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0106] As an example, the executable instructions may or may not correspond to a file in the file system, and may be stored as a part of a file that stores other programs or data. For example, they may be stored in one or more scripts in a Hyper Text Markup Language (HTML) document, stored in a single file dedicated to the program being discussed, or stored in multiple cooperating files (such as files that store one or more modules, subroutines, or code portions).

[0107] As an example, the executable instructions may be deployed to be executed on one computing device, or on multiple computing devices located at one location, or on multiple computing devices distributed at multiple locations and interconnected by a communication network.

[0108] It should be noted that, in this document, the terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or system comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or system. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or system comprising that element.

[0109] The serial numbers of the embodiments of the present application above are for description only and do not represent the superiority or inferiority of the embodiments.

[0110] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as a read-only memory / random access memory, magnetic disk, optical disk), and includes several instructions for causing a multimedia terminal device (which may be a mobile phone, a computer, a television receiver, or a network device, etc.) to execute the methods described in the various embodiments of the present application.

[0111] The above are only the preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the description of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A panel defect detection method based on multi-model fusion, characterized in that: The method comprises the following steps: Performing image smoothing, histogram equalization and contrast enhancement processing on the original panel image to obtain a panel sample image; The panel sample images are tested for primary defects by combining edge detection and isolation forest judgment to obtain defects of abnormal screen display categories; Use image classification model to perform secondary defect detection on panel sample images to obtain screen scratch category defects; The target detection model is used to finally perform three-level defect detection on the panel sample image to obtain the defects of the screen bad pixels and bad lines; The results of defects in the screen display abnormality category, screen scratch category, and screen bad pixel and bad line category are fused to obtain panel defect detection results.

2. The panel defect detection method based on multi-model fusion according to claim 1 is characterized in that: The process of performing primary defect detection on panel sample images by combining edge detection and isolation forest judgment is as follows: Perform edge contour detection on the panel sample image based on the edge detection algorithm to obtain the maximum circumscribed rectangle of the panel; The maximum circumscribed rectangle of the panel is taken as the region of interest, and whether the size of the region of interest is abnormal is determined based on the isolation forest algorithm. If there is an abnormality, it is determined that the panel has a screen display abnormality category defect; otherwise, it is determined that the panel does not have a screen display abnormality category defect.

3. The panel defect detection method based on multi-model fusion according to claim 1 is characterized in that: The process of using the image classification model to perform secondary defect detection on panel sample images is as follows: Input the panel sample image into the image classification model, perform defect classification detection and normalization processing on the panel sample image through the image classification model, and output the defect category probability of the screen scratch category; Based on the threshold verification method, it is determined whether the defect category probability of the screen scratch category meets the verification standard. If the verification standard is met, it is determined that the panel has a screen scratch category defect. Otherwise, it is determined that the panel does not have a screen scratch category defect.

4. The panel defect detection method based on multi-model fusion according to claim 3 is characterized in that: The structure of the image classification model includes a shallow convolutional neural network and a normalized activation function.

5. The panel defect detection method based on multi-model fusion according to claim 1 is characterized in that: The process of using the target detection model to perform three-level defect detection on panel sample images is as follows: Input the panel sample image into the target detection model, and use the target detection model to locate and classify the panel sample image to obtain the target candidate frame; The target candidate boxes are subjected to non-maximum suppression processing to finally obtain target candidate boxes with screen bad pixels and bad line category defects.

6. The panel defect detection method based on multi-model fusion according to claim 5 is characterized in that: The structure of the target detection model includes a deep convolutional neural network, a region proposal network and a non-maximum suppression function.

7. The panel defect detection method based on multi-model fusion according to claim 1 is characterized in that: The process of fusing the results of screen display abnormality defects, screen scratch defects, and screen bad pixel and bad line defects is as follows: Sorting different categories of defects of panel sample images according to the priority sorting rules of defective categories; The sorted defects of different categories are screened in combination with the confidence level to obtain the final panel defect detection result.

8. A panel defect detection system based on multi-model fusion, characterized in that: The system comprises: An image preprocessing unit, the image preprocessing unit is used to perform image smoothing, histogram equalization and contrast enhancement on the original panel image to obtain a panel sample image; A display anomaly detection unit, wherein the display anomaly detection unit performs primary defect detection on the panel sample image by combining edge detection and isolation forest judgment to obtain defects of the screen display anomaly category; A screen scratch detection unit, wherein the screen scratch detection unit uses an image classification model to perform secondary defect detection on a panel sample image to obtain defects of the screen scratch category; A bad pixel and bad line detection unit, wherein the bad pixel and bad line detection unit uses a target detection model to perform three-level defect detection on a panel sample image to obtain defects of the screen bad pixel and bad line category; The defect result fusion unit is used to fuse the results of defects in the screen display abnormality category, screen scratch category, and screen bad pixel and bad line category to obtain panel defect detection results.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for detecting panel defects based on multi-model fusion as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for detecting panel defects based on multi-model fusion as described in any one of claims 1 to 7 is implemented.

Images