Liquid crystal display panel visual inspection system and method

Abstract

The application relates to the technical field of liquid crystal display panel detection, and discloses a liquid crystal display panel visual detection system and method, which comprises the following steps: S1, acquiring real-time image information of a liquid crystal display panel and production parameters of a production line; S2, performing multi-dimensional analysis according to the real-time image information to acquire multi-dimensional evaluation indexes; and S3, establishing a panel qualification judgment model according to the production parameters, inputting the multi-dimensional evaluation indexes into the panel qualification judgment model, and obtaining a detection result. According to the multi-dimensional analysis of the liquid crystal display panel based on the real-time image information, the panel qualification judgment model is established according to the production parameters of the production line, the quality of the liquid crystal display panel is judged through the panel qualification judgment model and the multi-dimensional evaluation indexes, the 'apparent quality' and 'process health degree' of the panel can be comprehensively reflected, the misjudgment caused by single-dimensional information can be avoided, the organic fusion of multi-source heterogeneous information is realized, and the robustness of decision-making is improved.

Description

Technical Field

[0001] This application relates to the technical field of liquid crystal display panel inspection, and in particular to a visual inspection system and method for liquid crystal display panels. Background Technology

[0002] LCD panels are widely used in many fields such as consumer electronics, industrial control, and medical equipment. Their quality directly affects the display effect and user experience of end products. As display technology continues to evolve towards higher resolution, higher refresh rate, and narrower bezels, and as the market's expectations for zero-defect products increase, quality control in the LCD panel manufacturing process has become a key factor in determining the core competitiveness of products.

[0003] Traditional LCD panel defect detection primarily relies on manual visual inspection. Quality inspectors illuminate the panel in a dark room and switch between standard images such as red, green, blue, white, and black, judging the presence of defects based on visual experience. However, manual inspection has significant limitations; repetitive work over a long period leads to decreased attention, resulting in missed detections and misjudgments. With the application of AOI systems, LCD panel inspection technology has advanced. AOI systems utilize high-resolution cameras, dedicated optical illumination, and image processing software to acquire and analyze images of the panel under inspection, completing the inspection of multiple images in a short time, far exceeding the efficiency of manual inspection. However, traditional AOI systems still have limitations in detecting point defects and line defects. While performing well with high contrast and well-defined defects, it faces bottlenecks such as difficulty in feature extraction and weak generalization ability when dealing with low contrast, blurred edges, and Mura defects of various shapes. With the development of AI technology, AI vision-based automatic defect classification systems can automatically identify various micron-level defects on the production line. However, current detection methods still focus on the independent identification and detection of various defects, which is relatively independent of the judgment process of production parameter health. Furthermore, they lack the ability to detect the quality of LCD panels using human visual perception. Therefore, how to integrate multi-dimensional feature information and simulate human visual perception to detect the quality of LCD panels is the fundamental problem that this invention aims to solve. Summary of the Invention

[0004] In order to integrate multi-dimensional feature information and simulate human visual perception to detect the quality of liquid crystal panels, this application provides a visual inspection system and method for liquid crystal display panels.

[0005] In a first aspect, this application provides a visual inspection method for a liquid crystal display panel, employing the following technical solution:

[0006] A visual inspection method for liquid crystal display panels, comprising:

[0007] S1. Obtain real-time image information of the LCD panel and production parameters of the production line;

[0008] S2. Perform multi-dimensional analysis based on real-time image information to obtain multi-dimensional evaluation indicators;

[0009] S3. Establish a panel acceptance judgment model based on production parameters, input multi-dimensional evaluation indicators into the panel acceptance judgment model, and obtain the test results.

[0010] Optionally, the production parameters include multi-dimensional evaluation indicators, qualification status, and corresponding process parameters of historical liquid crystal display panels;

[0011] The process of establishing the panel qualification judgment model includes:

[0012] Obtain the average yield of the LCD panel within a preset time period;

[0013] The probability of this multi-dimensional evaluation index appearing in qualified and unqualified LCD panels is calculated separately.

[0014] Establish the probability distribution of each production parameter based on the data of qualified and unqualified LCD panels respectively, and obtain the probability of each production parameter in the qualified category and the probability of each production parameter in the unqualified category.

[0015] The likelihood is obtained by multiplying the probability of the appearance of multi-dimensional evaluation indicators in qualified LCD panels with the probability of each production parameter in the qualified category.

[0016] The probability of multi-dimensional evaluation indicators appearing in the defective LCD panel is multiplied by the probability of each production parameter of the defective class to obtain the defect likelihood.

[0017] The pass rate is obtained based on the average yield, likelihood, and defect likelihood. The pass rate is then compared with the decision threshold to determine whether the test is qualified.

[0018] Optionally, the multi-dimensional analysis process includes:

[0019] The visual saliency index and performance deviation index of the LCD panel are calculated based on real-time image information.

[0020] Identify defect areas from real-time image information and determine the damage index based on the defect areas;

[0021] Multidimensional evaluation indicators were determined based on the visual salience index, performance deviation index, and damage index.

[0022] Optionally, the calculation process of the visual saliency index includes:

[0023] Flatten the background signal of real-time image information;

[0024] The image is decomposed into different detail coefficients by wavelet transform, the detail coefficients are weighted based on the contrast sensitivity function, and the weighted detail coefficients are subjected to inverse discrete wavelet transform to obtain the adjusted local contrast.

[0025] The visual saliency weights are determined based on the brightness, color, orientation selectivity, and central-peripheral differences in real-time image information.

[0026] The visual salience index is calculated based on local contrast, visual salience weight, and region area.

[0027] Optionally, the process of determining the saliency weight includes:

[0028] Brightness characteristics are determined based on local brightness contrast.

[0029] Color features are determined by the contrastive color mechanism in the real-time optical information space (LAB).

[0030] Multi-directional energy is extracted based on Gabor filter banks, and the maximum response is used as a directional selectivity feature.

[0031] Using fine features as the center, the absolute difference between the center and the surrounding area is used as the center-surround difference map;

[0032] The center-periphery difference maps of brightness features, color features, and orientation selectivity features are weighted and summed, and then mapped using the Sigmoid function to obtain the saliency weights.

[0033] Optionally, the calculation process of the performance deviation index includes:

[0034] The brightness uniformity deviation is determined based on the maximum brightness deviation, the standard deviation of brightness deviation, and the integral of the brightness gradient amplitude in the real-time image information.

[0035] The color gamut coverage deviation is determined based on the color gamut area ratio, HD distance, and the color coordinate vectors of the three primary colors.

[0036] Calculate the response time deviation based on the response time, maximum absolute deviation, and overshoot.

[0037] The weighted sum of brightness uniformity deviation, color gamut coverage deviation, and response time deviation is used as the performance deviation index.

[0038] Optionally, the calculation process of the damage index includes:

[0039] The background texture of the liquid crystal display panel in real-time image information is removed by weighted template difference map method, and the defect area is obtained by maximum entropy threshold segmentation method.

[0040] The degree of damage to each curve region is determined based on geometric and visual properties.

[0041] Determine the distribution entropy based on the distribution status of the defect area;

[0042] The damage index is calculated based on the damage degree and distribution entropy of all defective regions.

[0043] Optionally, the calculation process of the multi-dimensional evaluation index includes:

[0044] Obtain the weighted sum of the visual saliency index, performance deviation index, and damage index, and use the sum of the weighted sum and the coupling term between any two of the visual saliency index, performance deviation index, and damage index as a multi-dimensional evaluation index.

[0045] Secondly, this application provides a visual inspection system for liquid crystal display panels, which adopts the following technical solution:

[0046] A visual inspection system for liquid crystal display panels, the system employing any one of the visual inspection methods for liquid crystal display panels described above, comprising an illumination system, an image processing module, and:

[0047] A camera used to acquire real-time image information from an LCD display panel;

[0048] The production line data interface module is used to obtain the production parameters of the LCD display panel;

[0049] The analysis model is used to perform multi-dimensional analysis based on real-time image information and obtain multi-dimensional evaluation indicators.

[0050] The panel acceptance judgment model is used to obtain test results based on multi-dimensional evaluation indicators.

[0051] In summary, this application includes at least one of the following beneficial technical effects:

[0052] This invention performs multi-dimensional analysis of liquid crystal display panels based on real-time image information, and also establishes a panel qualification judgment model according to the production parameters of the production line. By judging the quality of liquid crystal display panels through the panel qualification judgment model and multi-dimensional evaluation indicators, it can comprehensively reflect the "apparent quality" and "process health" of the panel, avoid misjudgment caused by single-dimensional information, realize the organic integration of multi-source heterogeneous information, and improve the robustness of decision-making. Attached Figure Description

[0053] Figure 1 This is a logical schematic diagram of the visual inspection method for liquid crystal display panels in this invention;

[0054] Figure 2 This is a logical schematic diagram of the visual inspection system for liquid crystal display panels in this invention. Detailed Implementation

[0055] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0056] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0057] This application discloses a visual inspection method for liquid crystal display panels, referring to... Figure 1 The process includes: S1, acquiring real-time image information of the liquid crystal display panel and production parameters of the production line; S2, performing multi-dimensional analysis based on the real-time image information to obtain multi-dimensional evaluation indicators; S3, establishing a panel qualification judgment model based on the production parameters, inputting the multi-dimensional evaluation indicators into the panel qualification judgment model, and obtaining the test results. It can be seen that this embodiment, while performing multi-dimensional analysis of the liquid crystal display panel based on real-time image information, also establishes a panel qualification judgment model based on the production line parameters. By judging the quality of the liquid crystal display panel through the panel qualification judgment model and multi-dimensional evaluation indicators, it can comprehensively reflect the panel's "apparent quality" and "process health," avoiding misjudgments caused by single-dimensional information, achieving the organic integration of multi-source heterogeneous information, and improving the robustness of decision-making.

[0058] In one embodiment, production parameters include multi-dimensional evaluation indicators, pass / fail status, and corresponding process parameters of historical liquid crystal display panels; the process of establishing the panel pass / fail judgment model includes: firstly, obtaining the average yield of liquid crystal display panels within a preset time period based on historical data from the production line. Therefore, the average defect rate is .

[0059] Then, the probability of the multi-dimensional evaluation index S appearing in both qualified and unqualified LCD panels was calculated. , Commonly used models include Gaussian distribution, beta distribution, log-normal distribution, etc. The model is selected to calculate the probability based on the distribution of the multi-dimensional evaluation index S. In this embodiment, Gaussian distribution is used to calculate the probability.

[0060] Each production parameter is established based on the data from both qualified and unqualified LCD panels. The probability distribution, obtaining the probability of each production parameter for the qualified class. and the probability of each production parameter in the non-conforming category. For example, a probability distribution is established based on the temperature of qualified glass substrates from historical qualified products before they enter the equipment, serving as a typical distribution. In this embodiment, the production parameters are all continuous parameters; therefore, a Gaussian distribution is used as the model for the typical distribution. If the current panel's production parameters... If it deviates significantly from the typical distribution of qualified products, then the corresponding The difference will be smaller, which will affect the subsequent judgment results.

[0061] Then, the probability of the appearance of multi-dimensional evaluation indicators in a qualified LCD panel is multiplied by the probability density of each production parameter in the corresponding probability to obtain the likelihood. ; Likelihood indicates the probability that the current panel possesses the defect score and production parameters within the group of qualified products.

[0062] The probability of multi-dimensional evaluation indicators appearing in defective LCD panels is multiplied by the probability of each production parameter in the defective category to obtain the defect likelihood. ; Defect likelihood indicates the probability that the current panel possesses the defect score and production parameters within the group of nonconforming products.

[0063] Finally, the pass / fail probability is obtained based on the average yield, likelihood, and defect likelihood. , Therefore, the probability of passing is expressed in the average yield. Based on this, the probability that the panel should be a qualified product is reassessed according to its specific defects (multi-dimensional evaluation indicators) and production data (process parameters). By comparing the qualification probability with a decision threshold, which is set according to the manufacturer's strictness in product quality control, the test result is judged as qualified when the qualification probability is greater than or equal to the decision threshold, and otherwise as unqualified. Through the above process, compared with the method of directly judging the threshold based on multi-dimensional evaluation indicators, it can simultaneously integrate visual evidence obtained from image analysis and process evidence collected on the production line, thereby improving the accuracy of defect analysis.

[0064] In one embodiment, a multi-dimensional analysis process is provided, which includes: calculating the visual saliency index of the liquid crystal display panel based on real-time image information. The calculation process includes: firstly, flattening the background signal of the real-time image information; this process is to remove large-scale global brightness variations, and fitting the brightness distribution of the background using a bivariate polynomial. Then, subtract it from the original image to obtain an image with a flat background, mathematically expressed as: , This is the image after the background signal has been flattened. For the original image, is the fitted bivariate polynomial.

[0065] Next, the image is decomposed into different detail coefficients using wavelet transform, the process of which is as follows:

[0066] ,in, These are approximation coefficients, which preserve the low-frequency contour information of the image. For detail coefficients, high-frequency details in the horizontal, vertical, and diagonal directions were preserved respectively; then, the detail coefficients were weighted based on the contrast sensitivity function, and... Updated to ,Will Updated to ,Will Updated to Where A is the amplitude scaling factor, β is the low-frequency rise exponent, and α is the high-frequency attenuation coefficient. A, β, and α are all constants in the contrast sensitivity function. The spatial frequencies of the HL, LH, and HH subbands are respectively; the weighted detail coefficients are subjected to inverse discrete wavelet transform to obtain the adjusted local contrast H(C(x,y),T):

[0067] Then, visual saliency weights are determined based on brightness, color, orientation selectivity, and central-peripheral differences in real-time image information. This process includes: first, determining brightness features based on local brightness contrast. ; , For (x,y) coordinates, the L (luminance) channel in LAB space. The mean value of the L-channels in the local neighborhood centered at coordinates (x,y). To prevent dividing by zero and positive numbers, Therefore when The larger the value, the higher the weight that defects in high-contrast areas can be assigned.

[0068] Secondly, color features are determined based on the contrastive color mechanism of the real-time optical information space (LAB). ; ,in, This indicates the degree to which the color of the (x,y) coordinate deviates from a neutral color along the red-green axis. This indicates the degree to which the color of the (x,y) coordinate deviates from neutral on the blue-yellow axis, through... Calculating visual saliency weights can increase the judgment weight of color defects at a given location when color saturation is high.

[0069] Subsequently, multi-directional energy is extracted based on the Gabor filter bank, and its maximum response is used as a direction selectivity feature. ; ,in, , For Gabor kernels, the parameters include As direction, For the scale, in this embodiment Set to 2, For wavelength, in this embodiment Set to 4; via direction selectivity feature The size of the value can simulate the sensitivity of the human visual system to edges, textures and lines in different directions. The larger the value, the higher the weight of defect judgment.

[0070] Then, using the fine features as the center, the absolute difference between the center and the surrounding area is used as the center-surround difference map. , Where s represents the central scale, s+2 represents the peripheral scale, s+2≤S, and S is the total number of pyramid layers.

[0071] Subsequently, the center-periphery difference maps of brightness features, color features, and orientation selectivity features are weighted and summed, and then mapped using the Sigmoid function to obtain the saliency weights. The resulting visual saliency weights are:

[0072] sig is the sigmoid function, used to map values ​​between 0 and 1; through the process of establishing the visual saliency weight W(x,y), the area that "human eyes really focus on" can be highlighted, the detection rate of low contrast Mura can be improved, and background texture and periodic noise can be suppressed, thus improving the consistency with human subjective rating.

[0073] Finally, the visual salience index is calculated based on local contrast, visual salience weight, and region area. The calculation model is as follows: ,in, The visual contrast threshold. The area of ​​the Mura (scar) region. As an area penalty factor, The total area is denoted as . It should be noted that the calculation process of the visual saliency index is based on the existing Mura defect detection algorithm. The improvement lies in the process of obtaining the adjusted local contrast and the introduction and acquisition of the visual saliency weight. Therefore, the calculation model will not be described in detail.

[0074] In one embodiment, the calculation process for the performance deviation index is given, which includes: the maximum brightness deviation in real-time image information. Standard deviation of brightness and the integral of the brightness gradient magnitude Determine the brightness uniformity deviation; the calculation process is as follows: based on the maximum brightness deviation... Standard deviation of brightness and the integral of the brightness gradient magnitude Obtaining brightness influence items , , , This is a standard brightness reference value. The spatial average value of the standard brightness is used as the weighted sum of the influencing factors as the brightness uniformity deviation D1. The brightness uniformity deviation D1 can be used to measure the brightness uniformity status from the perspectives of brightness deviation, global non-uniformity measurement, and local abrupt changes.

[0075] Then, based on the color gamut area ratio HD distance and the color coordinate vectors of the three primary colors Determine the color gamut coverage deviation; where HD distance is the Hausdorff distance, which measures the measured color gamut boundary. boundary with standard color gamut Due to shape differences, the color influence term is obtained based on the color gamut area ratio, HD distance, and the color coordinate vectors of the three primary colors. , , , Represents Euclidean distance. The maximum possible color difference after normalization. The standard color coordinates of the three primary colors are used; therefore, the weighted sum of the color influence terms is used as the color gamut coverage deviation D2. By measuring the magnitude of the color gamut coverage deviation D2, the color gamut coverage deviation status can be determined from the color gamut area, color gamut shape similarity, and primary color deviation.

[0076] Then, based on the response time Maximum absolute deviation Calculate the response time deviation using the overshoot os(t); obtain the response time influence term. , , ,in, This represents the weight from gray level i to gray level j, which is set based on the common gray level transition probabilities in natural images. This represents the response time from gray level i to gray level j. This represents the standard time from gray level i to gray level j. The reference time constant is set according to the "maximum allowable response time" in industry standards or product specifications. By using the weighted sum of the response time influencing factors as the response time deviation D3, the deviation of the response time can be judged from the weighted average of different grayscale conversions and the overshoot and undershoot effects.

[0077] It should be noted that the weighted values ​​in the calculation of the above-mentioned brightness uniformity deviation D1, color gamut coverage deviation D2 and response time deviation D3 are set according to experimental data, and the sum of the weights involved in the calculation of each deviation is 1.

[0078] Finally, the weighted sum of brightness uniformity deviation, color gamut coverage deviation, and response time deviation is used as the performance deviation index, with the corresponding weighted sum being 1. The weights are set according to the product's purpose. For example, for home office LCD screens, the weights of brightness uniformity deviation, color gamut coverage deviation, and response time deviation are 0.5, 0.25, and 0.25, respectively. If the application is in the e-sports gaming field, the weight of response time deviation is increased. This will not be elaborated here.

[0079] In one embodiment, defect areas are also identified from real-time image information, and a damage index is determined based on the defect areas. The calculation process of the damage index includes: removing the background texture of the liquid crystal display panel from the real-time image information based on the weighted template difference map method. Since the periodic pixel structure of the liquid crystal display panel itself is the background texture, the "weighted template difference map method" can effectively deduct this background and highlight isolated point or line defects; and obtaining the defect area based on the maximum entropy threshold segmentation method.

[0080] The damage degree of each curve region is then determined based on geometric and visual attributes. Geometric attributes include point defects, line defects, and region defects; visual attributes are obtained based on the contrast and positional weights of the defects. Therefore, the damage degree calculation model is as follows: Where N is the normalization factor, which is the same as ensuring The values ​​fall within a reasonable preset range; , This is a control factor used to control the contribution of each factor in the nonlinear combination. In this embodiment... , Set all to 1.

[0081] For geometric attribute factors, when it is a point defect, , The pixel area occupied by a point defect. The area of ​​a single pixel.

[0082] When it is a line defect , For line length, For line width, For reference line length, The width is in pixels. , These are nonlinear factors for length and width, respectively, which can be extracted from the background using methods such as weighted matrix decomposition models. In this embodiment, they are both set to 1.

[0083] When it is a regional defect , The area of ​​damage. This is the area penalty factor, which is set based on empirical data and has a value greater than 0. For shape complexity factor, , Perimeter of the defective area.

[0084] The visual attribute factor is calculated as follows:

[0085]

[0086]

[0087] in, Let be the contrast of the k-th defect. This is a visual nonlinear factor, obtained through data simulation. for Position weight, It is a fixed coefficient, with a value range of 0 to 2, and in this embodiment, its value is 0.8; Let the center coordinates of the k-th defect be... The coordinates are the center coordinates of the screen.

[0088] Then, the distribution entropy is determined based on the distribution status of the defect area. This process first clusters spatially similar defects into Given clusters, obtain the probability that each defect belongs to the y-th cluster. After that, through Calculate the distribution entropy .

[0089] Finally, the damage index is calculated based on the damage degree and distribution entropy of all defective regions. The calculation process is as follows: ,in, Let be the aggregation penalty intensity coefficient, when When the value is 0, there is no penalty for gathering. When the value is 1, there is a slight penalty; therefore, The value is set according to the application scenario of the LCD panel. For general consumer-grade panels, the value setting range is 1.5~2.0, and for medical and automotive display panels, the value setting range is 2.5~3.

[0090] After obtaining the visual saliency index, performance deviation index, and damage index, a multi-dimensional evaluation index is determined by combining the above indices.

[0091] In one embodiment, the weighted sum of the visual saliency index, performance deviation index, and damage index is used as a multi-dimensional evaluation index, and the weight of each index is adaptively changed according to the severity of that dimension.

[0092] In another embodiment, the process of calculating the multi-dimensional evaluation index first obtains the weighted sum of the visual saliency index, performance deviation index, and damage index, and then obtains the coupling term between any two of the visual saliency index, performance deviation index, and damage index. The value of the coupling term is set according to the category of the corresponding index and empirical data. When two types of defects occur at the same time, the value of the multi-dimensional evaluation index is increased by the coupling term. The larger the value of the multi-dimensional evaluation index, the more serious the defect. Therefore, using the weighted sum and the sum of the coupling term between any two of the visual saliency index, performance deviation index, and damage index as the multi-dimensional evaluation index can identify the nonlinear superposition effect of the overall quality.

[0093] This application also discloses a visual inspection system for liquid crystal display panels, which is described in any of the above-mentioned methods for visual inspection of liquid crystal display panels. Please refer to the appendix. Figure 2 The system includes a lighting system, an image processing module, and: a camera for acquiring real-time image information of the LCD panel; a production line data interface module for acquiring production parameters of the LCD panel; an analysis model for performing multi-dimensional analysis based on real-time image information to obtain multi-dimensional evaluation indicators; and a panel qualification judgment model for obtaining test results based on multi-dimensional evaluation indicators.

[0094] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A visual inspection method for a liquid crystal display panel, characterized in that, include: S1. Obtain real-time image information of the LCD panel and production parameters of the production line; S2. Perform multi-dimensional analysis based on real-time image information to obtain multi-dimensional evaluation indicators; S3. Establish a panel acceptance judgment model based on production parameters, input multi-dimensional evaluation indicators into the panel acceptance judgment model, and obtain the test results.

2. The visual inspection method for a liquid crystal display panel according to claim 1, characterized in that, The production parameters include multi-dimensional evaluation indicators, qualification status and corresponding process parameters of historical liquid crystal display panels. The process of establishing the panel qualification judgment model includes: Obtain the average yield of the LCD panel within a preset time period; The probability of this multi-dimensional evaluation index appearing in qualified and unqualified LCD panels is calculated separately. Establish the probability distribution of each production parameter based on the data of qualified and unqualified LCD panels respectively, and obtain the probability of each production parameter in the qualified category and the probability of each production parameter in the unqualified category. The likelihood is obtained by multiplying the probability of the appearance of multi-dimensional evaluation indicators in qualified LCD panels with the probability of each production parameter in the qualified category. The probability of multi-dimensional evaluation indicators appearing in the defective LCD panel is multiplied by the probability of each production parameter of the defective class to obtain the defect likelihood. The pass rate is obtained based on the average yield, likelihood, and defect likelihood. The pass rate is then compared with the decision threshold to determine whether the test is qualified.

3. The visual inspection method for a liquid crystal display panel according to claim 1, characterized in that, The multidimensional analysis process includes: The visual saliency index and performance deviation index of the LCD panel are calculated based on real-time image information. Identify defect areas from real-time image information and determine the damage index based on the defect areas; Multidimensional evaluation indicators were determined based on the visual salience index, performance deviation index, and damage index.

4. The visual inspection method for a liquid crystal display panel according to claim 3, characterized in that, The calculation process of the visual saliency index includes: Flatten the background signal of real-time image information; The image is decomposed into different detail coefficients by wavelet transform, the detail coefficients are weighted based on the contrast sensitivity function, and the weighted detail coefficients are subjected to inverse discrete wavelet transform to obtain the adjusted local contrast. The visual saliency weights are determined based on the brightness, color, orientation selectivity, and central-peripheral differences in real-time image information. The visual salience index is calculated based on local contrast, visual salience weight, and region area.

5. The visual inspection method for a liquid crystal display panel according to claim 4, characterized in that, The process of determining the significance weight includes: Brightness characteristics are determined based on local brightness contrast. Color features are determined by the contrastive color mechanism in the real-time optical information space (LAB). Multi-directional energy is extracted based on Gabor filter banks, and the maximum response is used as a directional selectivity feature. Using fine features as the center, the absolute difference between the center and the surrounding area is used as the center-surround difference map; The center-periphery difference maps of brightness features, color features, and orientation selectivity features are weighted and summed, and then mapped using the Sigmoid function to obtain the saliency weights.

6. The visual inspection method for a liquid crystal display panel according to claim 3, characterized in that, The calculation process for the performance deviation index includes: The brightness uniformity deviation is determined based on the maximum brightness deviation, the standard deviation of brightness deviation, and the integral of the brightness gradient amplitude in the real-time image information. The color gamut coverage deviation is determined based on the color gamut area ratio, HD distance, and the color coordinate vectors of the three primary colors. Calculate the response time deviation based on the response time, maximum absolute deviation, and overshoot. The weighted sum of brightness uniformity deviation, color gamut coverage deviation, and response time deviation is used as the performance deviation index.

7. The visual inspection method for a liquid crystal display panel according to claim 3, characterized in that, The calculation process of the damage index includes: The background texture of the liquid crystal display panel in real-time image information is removed by weighted template difference map method, and the defect area is obtained by maximum entropy threshold segmentation method. The degree of damage to each curve region is determined based on geometric and visual properties. Determine the distribution entropy based on the distribution status of the defect area; The damage index is calculated based on the damage degree and distribution entropy of all defective regions.

8. The visual inspection method for a liquid crystal display panel according to claim 3, characterized in that, The calculation process for the multi-dimensional evaluation indicators includes: Obtain the weighted sum of the visual saliency index, performance deviation index, and damage index, and use the sum of the weighted sum and the coupling term between any two of the visual saliency index, performance deviation index, and damage index as a multi-dimensional evaluation index.

9. A visual inspection system for liquid crystal display panels, characterized in that, The system employs a visual inspection method for liquid crystal display panels as described in any one of claims 1-8, including an illumination system, an image processing module, and: A camera used to acquire real-time image information from an LCD display panel; The production line data interface module is used to obtain the production parameters of the LCD display panel; The analysis model is used to perform multi-dimensional analysis based on real-time image information and obtain multi-dimensional evaluation indicators. The panel acceptance judgment model is used to obtain test results based on multi-dimensional evaluation indicators.

Images