Liquid crystal substrate glass size measurement method based on artificial intelligence and image analysis

Abstract

The invention discloses a liquid crystal substrate glass size measurement method based on artificial intelligence and image analysis, and relates to the technical field of glass measurement, and the method comprises the following steps: I, sequentially triggering sensors at different positions in the conveying process of liquid crystal substrate glass, and dynamically adjusting the conveying speed according to the triggering condition of the sensors until the liquid crystal substrate glass reaches a measurement position; according to the method, the influence of noise, uneven illumination and glass micro-warping on a measurement result is effectively reduced, the size measurement speed and precision are remarkably improved, background change, environmental noise, reflection interference and illumination fluctuation in a complex scene can be effectively dealt with, the stability of the measurement result under different working conditions is ensured, continuous and online high-speed measurement is realized, and the measurement precision is improved. The real-time detection requirement of a production line is met, the detection speed and the size resolution capability are improved on the premise that the hardware cost is not increased, systematic errors and long-term drift can be automatically corrected, the measurement consistency and reliability are improved, frequent manual adjustment and calibration are avoided, and the maintenance cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of glass measurement technology, and more specifically to a method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis. Background Technology

[0002] The dimensions of a liquid crystal substrate glass refer to the dimensions of its four sides. Currently, there are two main non-contact measurement methods: one is to use a line scan camera to scan line by line, then stitch the images together, and finally calculate the dimensions based on the images; the other is to use a high-precision industrial camera, calibrate it, take pictures, and then calculate the dimensions using image processing technology. The main drawback of the first method is its long processing time, and the main drawback of the second method is its low accuracy. This technology combines artificial intelligence and image analysis technology, avoiding the above two drawbacks and providing a fast and highly accurate measurement method. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies and provide a method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis.

[0004] This invention proposes a method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis, comprising the following steps: S1. The camera is calibrated using a standard measuring plate, the acquired glass image is preprocessed, and the glass contour is extracted. At the same time, the edge geometry of the glass is compensated and corrected, and an affine matrix M is established. S2. During the transfer process, the liquid crystal substrate glass triggers sensors at different positions in sequence. Based on the sensor triggering status, the transfer speed is dynamically adjusted until the measurement position is reached. S3. After the LCD substrate glass is in place, the orientation of the LCD substrate glass is corrected. After the glass orientation is stable, a photo capture command is sent, and the camera captures the glass image according to the set parameters. S4. Perform image analysis on the acquired images and convert the glass contour information in the image coordinate system into the actual physical dimensions in the measurement coordinate system through affine transformation to obtain the initial dimensions of the glass. S5. Using the established intelligent prediction model, the initial glass size is taken as input, and the initial glass size is adjusted through the intelligent prediction model. S6. Output the adjusted glass size data and transfer the glass to the next process. Based on actual needs, decide whether to add the glass to the historical training set to achieve parameter self-optimization adjustment.

[0005] As a further aspect of the present invention, the camera calibration using a standard measuring plate in step S1 specifically involves: after the camera position and field of view are fixed, aligning the standard measuring plate with the corner points and placing it close to the positioning block, acquiring an image of the standard measuring plate, then filtering the image to obtain the precise contour of the standard measuring plate image, analyzing the affine relationship of the image based on the standard size of the measuring plate, and calculating the corresponding affine matrix M; simultaneously, calculating the camera resolution R based on the correspondence between the image coordinate system and the measurement coordinate system.

[0006] In addition, it should be noted that camera precision = camera single-sided pixels / actual size of the corresponding side (mm); standard measuring plate single-sided pixels = measuring plate single-sided size (mm) * camera precision; affine matrix M = standard measuring plate actual size (pixel) matrix / acquired image matrix.

[0007] As a further aspect of the present invention, the specific steps for preprocessing the acquired glass image in step S1 are as follows: S1.1: The color image of the liquid crystal substrate glass is acquired by the camera. Then, based on the contribution of the RGB three channels to the contrast of the glass edge, the color image is converted into a single-channel grayscale image. The grayscale value of each pixel in the grayscale image is compared with the preset grayscale threshold. If the grayscale value is lower than the grayscale threshold, the corresponding pixel area is determined to be the background area. Otherwise, it is the glass area, and the identified background area is suppressed. S1.2: Morphological opening operation is used to first erode the grayscale image, then dilate to restore the main structure of the glass, remove small-scale background noise in the grayscale image, and then dilate and erode the grayscale image to fill the internal holes in the grayscale image. S1.3: Sort the gray values ​​in the neighborhood of each pixel in the grayscale image from high to low, select the median value after sorting, perform median filtering on each pixel value in the grayscale image, and then perform Gaussian smoothing on the grayscale image after median filtering.

[0008] As a further aspect of the present invention, the specific conversion formula for the grayscale image in S1.1 is as follows: ; In the formula, Representing pixels The intensity of the red channel at that location; Representing pixels The intensity of the green channel at that location; Representing pixels The intensity of the blue channel at that location; , as well as These represent the weighting coefficients for each of the R, G, and B channels, respectively.

[0009] As a further aspect of the present invention, the specific steps for extracting the glass contour in step S1, and simultaneously compensating and correcting the edge geometry of the glass, are as follows: S2.1: At the same measurement location, acquire multiple modal images, and through time synchronization and spatial registration, map the images of different modalities to a unified coordinate system to establish corresponding joint feature maps, and then preprocess the established joint feature maps; S2.2: Calculate the spatial gradient change of the preprocessed joint feature map. If the spatial gradient change is higher than the preset change threshold, the corresponding region is marked as a gray-scale change region, and each pixel in the gray-scale change region is used as the initial candidate pixel of the glass edge. S2.3: Based on the preset fitting range, perform continuous function fitting on each group of initial candidate pixels and the pixels within the fitting range, and obtain the position of sub-pixel edges in each modal image by interpolation. Perform consistency judgment on sub-pixel edge points from different modalities. If the difference between two groups of sub-pixel edge points is less than the preset consistency threshold, then treat them as valid edge points and retain them. S2.4: Based on the camera's geometric imaging relationship, the retained sub-pixel edge points are mapped to three-dimensional space to obtain the three-dimensional coordinates of the glass edge in real space. Then, the spatial distribution of the three-dimensional edge points is fitted to perform geometric compensation and correction on the two-dimensional projected edge. After that, the compensated three-dimensional edge points are reprojected onto the measurement plane to generate the corresponding glass contour curve and record the glass contour points.

[0010] As a further aspect of the present invention, the multiple modal images mentioned in S2.1 specifically include visible light intensity maps, phase response maps, height response maps, etc.

[0011] As a further aspect of the present invention, the specific steps for obtaining the initial dimensions of the glass in step IV are as follows: S3.1: Classify and sort the glass contour points according to their geometric positions. Based on the classification and sorting results, divide each glass contour point into a set of points on the four sides: top, bottom, left, and right. Represent these points as two-dimensional points in the image coordinate system. Then, based on the camera calibration results, establish an affine mapping relationship from the image coordinate system to the physical measurement plane coordinate system. S3.2: Based on the current affine matrix, calculate the corresponding coordinates of the image coordinates of each contour point in the measurement plane. Then, perform line fitting on the physical coordinate contour points corresponding to each glass edge to obtain the geometric expression of the corresponding glass edge in physical space. S3.3: On each fitted straight line, based on the extreme position of the contour point in that direction, determine the two physical endpoints of the corresponding glass edge. Then, based on the two determined physical endpoints, calculate the straight-line distance of the corresponding glass edge in physical space to obtain the initial length value of each glass edge. Statistically calculate the obtained initial length values ​​of each glass edge to generate the corresponding initial glass size.

[0012] As a further aspect of the present invention, the specific steps for adjusting the initial glass size using the intelligent prediction model in step S5 are as follows: S4.1: Select multiple sets of glass samples with known true dimensions from the historical measurement process, and obtain the corresponding initial side length data of each set of glass samples from the glass images and geometric calculations. Then, match each set of side length data with the true dimensions measured manually or by high-precision instruments to construct a historical sample dataset, and then establish a size correction model. S4.2: The initial side length data are used as inputs and sequentially imported into the size correction model. The size correction model performs multi-parameter mapping on the initial side length data, outputs the predicted side length size, compares the predicted side length size with the corresponding real size, and calculates the prediction error of each historical sample. S4.3: Calculate the prediction error of each historical sample to obtain the total loss error of the size correction model. Minimize the total loss error of the size correction model, adjust the model parameters, and iteratively train the size correction model repeatedly until the total loss error converges to the preset range or the number of iterations is reached, then stop training. S4.4: Input the initial dimensions obtained from online measurement into the optimized dimension correction model, calculate the corrected final dimension result, and perform statistical analysis on the prediction results in real time to determine whether the error of the dimension correction model on the new sample is within the preset allowable range. If it exceeds the allowable range, the dimension correction model is retrained.

[0013] As a further aspect of the present invention, the specific calculation formula for the predicted side length dimension in S4.2 is as follows: ; In the formula, Representing the The predicted side length of a glass sample; Represents weight; Representative deviation; The specific formula for calculating the total loss error mentioned in S4.2 is as follows: ; In the formula, This represents the total loss error function. To minimize the total loss error, we need to solve for... ; Represents the number of glass samples; Representing the The actual size of the glass sample.

[0014] The beneficial effects of this invention are: This invention acquires color images of a liquid crystal substrate glass using a camera, converts them into grayscale images based on the contribution of the RGB three channels to edge contrast, and distinguishes between glass and background areas using grayscale thresholding. Background suppression is then applied, followed by morphological opening operations to remove small-scale background noise, closing operations to fill internal holes, and median filtering and Gaussian smoothing to reduce noise and grayscale fluctuations. Multimodal images are then acquired and time-synchronized and spatially registered to construct a joint feature map. Gradient analysis is used to extract initial candidate pixels for glass edges, and continuous function fitting and interpolation are used for sub-pixel edge localization. Consistency screening of sub-pixel edges under different modalities is performed to retain valid edge points. Three-dimensional edge coordinates are reconstructed based on imaging geometry, and geometric compensation and correction are applied to generate the glass contour. The contour points are then classified and sorted, and the image coordinates are transformed using affine mapping. Using physical coordinates, the system fits straight lines to each side and calculates the distance between endpoints to obtain the initial dimensions of the four sides of the glass. Finally, a dimension correction model is established based on historical samples. The initial dimensions are trained and optimized to minimize errors. The optimized dimension correction model is then used to correct the glass side length online, effectively reducing the impact of noise, uneven illumination, and glass micro-warping on the measurement results. This significantly improves the accuracy of dimension calculation and can effectively cope with background changes, reflection interference, and illumination fluctuations in complex environments. It ensures the stability of measurement results under different working conditions, enabling continuous, online, high-speed measurement to meet the real-time inspection needs of the production line. Without increasing hardware costs, it improves measurement speed and dimension resolution, automatically corrects systematic errors and long-term drift, improves measurement consistency and reliability, avoids frequent manual adjustments, and reduces maintenance costs. Attached Figure Description

[0015] The present invention will now be further described with reference to the accompanying drawings.

[0016] Figure 1 This is a block diagram of a method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis. Figure 2 This is a flowchart of image processing and analysis for a liquid crystal substrate glass size measurement method based on artificial intelligence and image analysis. Figure 3 This is a flowchart illustrating a method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis. Figure 4 This is a schematic diagram of the support platform for a liquid crystal substrate glass size measurement method based on artificial intelligence and image analysis. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides a method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis. See also... Figure 1-4 , Figure 1 A framework diagram of a liquid crystal substrate glass size measurement method based on artificial intelligence and image analysis provided in an embodiment of the present invention. The method includes the following steps: During the transport of the liquid crystal substrate glass, sensors at different locations are triggered sequentially. The transport speed is dynamically adjusted according to the sensor triggering status until the measurement position is reached.

[0020] In this embodiment, reference Figure 4 As can be seen, initially, the glass enters the carrying platform from the previous node. When the glass travels to the position of the infrared beam sensor F1, sensor F1 is triggered. When sensor F1 is triggered, the control system starts the transmission rod B. The transmission rod B drives the glass to continue moving forward at a preset speed. When the glass travels to the position of the infrared beam sensor F2, sensor F2 is triggered. When sensor F2 is triggered, it indicates that the glass is close to the measurement position. At this time, the speed of transmission rod B is adjusted to slowly transport the glass to the position of the infrared beam sensor F3. Sensor F3 is triggered. When sensor F3 is triggered, transmission rod B is stopped, and the glass stops at the measurement position.

[0021] After the LCD substrate glass is in place, its orientation is corrected. Once the glass orientation is stable, a photo capture command is sent, and the camera captures an image of the glass according to the set parameters.

[0022] It should be further explained that the reference Figure 4 As can be seen, after the glass is in place, it is lifted by the ball E, and at the same time, push rods G1 and G2 extend to push the glass to a position close to positioning blocks D1 and D2. After the glass is close to the positioning blocks, push rods G1 and G2 reset, and at the same time, ball E resets, and the glass is flattened. After the image acquisition is completed, the transmission rod B is started to transport the glass to the next node at a preset speed. When the glass leaves the position of the infrared beam sensor F3, the transmission rod B is stopped.

[0023] Camera calibration is performed using a standard gauge, the acquired glass images are preprocessed, and the glass contour is extracted. At the same time, the geometric information of the glass edges is compensated and corrected.

[0024] Specifically, a color image of the liquid crystal substrate glass is acquired by a camera. Then, based on the contribution of the RGB three channels to the contrast of the glass edge, the color image is converted into a single-channel grayscale image. The grayscale value of each pixel in the grayscale image is then compared with a preset grayscale threshold. If the grayscale value is lower than the grayscale threshold, the corresponding pixel area is determined to be a background area; otherwise, it is a glass area. The identified background area is then suppressed. Morphological opening operation is used to first erode the grayscale image, and then dilation is used to restore the main structure of the glass and remove small-scale background noise in the grayscale image. Then, closing operation is used to first dilate and then erode the grayscale image to fill the internal holes in the grayscale image. The grayscale values ​​in the neighborhood of each pixel in the grayscale image are sorted from high to low, and the median value of the sorted image is selected to perform median filtering on the pixel values ​​of the grayscale image. Finally, Gaussian smoothing is applied to the median-filtered grayscale image.

[0025] Specifically, at the same measurement location, multiple modal images are acquired, and through time synchronization and spatial registration, the images of different modalities are mapped to a unified coordinate system to establish corresponding joint feature maps. The established joint feature maps are then preprocessed, and the spatial gradient change of the preprocessed joint feature maps is calculated. If the spatial gradient change exceeds a preset change threshold, the corresponding region is marked as a gray-level abrupt change region. Simultaneously, each pixel in the gray-level abrupt change region is used as an initial candidate pixel for the glass edge. Based on a preset fitting range, continuous function fitting is performed on each group of initial candidate pixels and the pixels within their fitting range, and interpolation is then applied. The positions of sub-pixel edges in each modal image are obtained. The consistency of sub-pixel edge points from different modalities is judged. If the difference between two sets of sub-pixel edge points is less than a preset consistency threshold, they are regarded as valid edge points and retained. Based on the camera's geometric imaging relationship, the retained sets of sub-pixel edge points are mapped to three-dimensional space to obtain the three-dimensional coordinates of the glass edge in real space. Then, the spatial distribution of the three-dimensional edge points is fitted to perform geometric compensation and correction on the two-dimensional projected edge. After that, the compensated three-dimensional edge points are reprojected onto the measurement plane to generate the corresponding glass contour curve and record the glass contour points.

[0026] In this embodiment, after the camera position is adjusted and fixed, the standard measuring plate is aligned with the corner points and placed close to the positioning block. At this time, the standard measuring plate image is acquired, and then the standard measuring plate image is filtered to obtain the accurate outline of the standard measuring plate image. Based on the camera accuracy, the length and width dimensions of the standard measuring plate are converted into pixel dimensions. Then, based on the actual size matrix of the standard measuring plate and the pixel dimensions of the acquired image of the standard measuring plate, the corresponding affine matrix is ​​calculated.

[0027] In addition, it should be noted that camera precision = camera single-sided pixels / actual size of the corresponding side (mm); standard measuring plate single-sided pixels = measuring plate single-sided size (mm) * camera precision; affine matrix M = standard measuring plate actual size (pixel) matrix / acquired image matrix; The various modal images specifically include visible light intensity maps, phase response maps, and height response maps.

[0028] Based on the glass contour information, the glass contour information in image coordinates is converted into actual physical dimensions to obtain the initial dimensions of the glass.

[0029] Specifically, the glass contour points are sorted counterclockwise according to their geometric coordinates, and the glass contour points are divided into point sets and vertex sets for four sides: top, bottom, left, and right. These are represented as two-dimensional points in the image coordinate system. Then, based on the camera calibration results, an affine mapping relationship is established from the image coordinate system to the physical measurement plane coordinate system. Based on the current affine matrix, the corresponding coordinates of the image coordinates of each contour point in the measurement plane are calculated point by point. Then, a straight line is fitted to the physical coordinate contour points corresponding to each glass edge to obtain the geometric expression of the corresponding glass edge in physical space. On each fitted straight line, the two physical endpoints of the corresponding glass edge are determined according to the extreme position of the contour point in that direction. Then, based on the two determined physical endpoints, the straight-line distance of the corresponding glass edge in physical space is calculated to obtain the initial length value of each glass edge. The obtained initial length values ​​of each glass edge are statistically analyzed to generate the corresponding initial glass dimensions.

[0030] An intelligent prediction model is established based on historical sample data, and the initial size of the glass is used as input to adjust the initial size of the glass.

[0031] Specifically, multiple sets of glass samples with known true dimensions are selected from historical measurements. Simultaneously, the initial side length data corresponding to each set of glass samples is obtained from the glass images and geometric calculations. Each set of side length data is then mapped one-to-one with the true dimensions measured manually or by high-precision instruments to construct a historical sample dataset. A size correction model is then established, with each initial side length data used as input. The size correction model performs multi-parameter mapping on each initial side length data, outputting predicted side length dimensions. The predicted side length dimensions are compared with the corresponding true dimensions to calculate the prediction error for each historical sample. The prediction errors for each historical sample are statistically analyzed to obtain the total loss error of the size correction model. The total loss error of the size correction model is minimized, and the model parameters are adjusted. The size correction model is iteratively trained repeatedly until the total loss error converges to a preset range or the number of iterations is reached. Training is then stopped, and the initial dimensions obtained from online measurements are input into the optimized size correction model to calculate the corrected final size result. The prediction results are statistically analyzed in real time to determine whether the error of the size correction model on new samples remains within a preset allowable range. If it exceeds the allowable range, the size correction model is retrained.

[0032] In addition, it should be noted that the specific formula for calculating the predicted side length is as follows: ; In the formula, Representing the The predicted side length of a glass sample; Represents weight; Representative deviation; The specific formula for calculating the total loss error is as follows: ; In the formula, This represents the total loss error function. To minimize the total loss error, we need to solve for... ; Represents the number of glass samples; Representing the The actual size of the glass sample.

[0033] The system outputs the adjusted glass size data and transfers the glass to the next process. Simultaneously, during continuous measurement, the measurement parameters are self-optimized and adjusted in real time.

[0034] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for measuring the dimensions of liquid crystal substrate glass based on artificial intelligence and image analysis, characterized in that, Includes the following steps: S1. The camera is calibrated using a standard measuring plate, the acquired glass image is preprocessed, and the glass contour is extracted. At the same time, the edge geometry of the glass is compensated and corrected, and an affine matrix M is established. S2. During the transfer process, the liquid crystal substrate glass triggers sensors at different positions in sequence. Based on the sensor triggering status, the transfer speed is dynamically adjusted until the measurement position is reached. S3. After the LCD substrate glass is in place, the orientation of the LCD substrate glass is corrected. After the glass orientation is stable, a photo capture command is sent, and the camera captures the glass image according to the set parameters. S4. Perform image analysis on the acquired images and convert the glass contour information in the image coordinate system into the actual physical dimensions in the measurement coordinate system through affine transformation to obtain the initial dimensions of the glass. S5. Using the established intelligent prediction model, the initial glass size is taken as input, and the initial glass size is adjusted through the intelligent prediction model. S6. Output the adjusted glass size data and transfer the glass to the next process. Based on actual needs, decide whether to add the glass to the historical training set to achieve parameter self-optimization adjustment.

2. The method for measuring the size of liquid crystal substrate glass based on artificial intelligence and image analysis according to claim 1, characterized in that, The specific steps of camera calibration using a standard measuring plate in step S1 are as follows: After the camera position and field of view are adjusted and fixed, the standard measuring plate is aligned with the corner points and placed close to the positioning block. At this time, the standard measuring plate image is acquired, and then the standard measuring plate image is filtered to obtain the accurate contour of the standard measuring plate image. Combined with the standard size of the measuring plate, the affine relationship of the image is analyzed, and the corresponding affine matrix M is calculated. At the same time, the camera resolution R is calculated according to the correspondence between the image coordinate system and the measurement coordinate system.

3. The method for measuring the size of liquid crystal substrate glass based on artificial intelligence and image analysis according to claim 1, characterized in that, The specific steps for preprocessing the acquired glass image in step S1 are as follows: S1.1: The color image of the liquid crystal substrate glass is acquired by the camera. Then, based on the contribution of the RGB three channels to the contrast of the glass edge, the color image is converted into a single-channel grayscale image. The grayscale value of each pixel in the grayscale image is compared with the preset grayscale threshold. If the grayscale value is lower than the grayscale threshold, the corresponding pixel area is determined to be the background area. Otherwise, it is the glass area, and the identified background area is suppressed. S1.2: Morphological opening operation is used to first erode the grayscale image, and then the glass body structure is restored by dilation to remove small-scale background noise in the grayscale image. Then, the grayscale image is first dilated and then eroded by closing operation to fill the internal holes in the grayscale image. S1.3: Sort the gray values ​​in the neighborhood of each pixel in the grayscale image from high to low, select the median value after sorting, perform median filtering on each pixel value in the grayscale image, and then perform Gaussian smoothing on the grayscale image after median filtering.

4. The method for measuring the size of liquid crystal substrate glass based on artificial intelligence and image analysis according to claim 3, characterized in that, The specific steps for extracting the glass contour and compensating and correcting the edge geometry of the glass in step S1 are as follows: S2.1: At the same measurement location, acquire multiple modal images, and through time synchronization and spatial registration, map the images of different modalities to a unified coordinate system to establish corresponding joint feature maps, and then preprocess the established joint feature maps; S2.2: Calculate the spatial gradient change of the preprocessed joint feature map. If the spatial gradient change is higher than the preset change threshold, the corresponding region is marked as a gray-scale change region, and each pixel in the gray-scale change region is used as the initial candidate pixel of the glass edge. S2.3: Based on the preset fitting range, perform continuous function fitting on each group of initial candidate pixels and the pixels within the fitting range, and obtain the position of sub-pixel edges in each modal image by interpolation. Perform consistency judgment on sub-pixel edge points from different modalities. If the difference between two groups of sub-pixel edge points is less than the preset consistency threshold, then treat them as valid edge points and retain them. S2.4: Based on the camera's geometric imaging relationship, the retained sub-pixel edge points are mapped to three-dimensional space to obtain the three-dimensional coordinates of the glass edge in real space. Then, the spatial distribution of the three-dimensional edge points is fitted to perform geometric compensation and correction on the two-dimensional projected edge. After that, the compensated three-dimensional edge points are reprojected onto the measurement plane to generate the corresponding glass contour curve and record the glass contour points.

5. The method for measuring the size of liquid crystal substrate glass based on artificial intelligence and image analysis according to claim 4, characterized in that, The specific steps for obtaining the initial dimensions of the glass in step S4 are as follows: S3.1: Classify and sort the glass contour points according to their geometric positions. Based on the classification and sorting results, divide each glass contour point into a set of points on the four sides: top, bottom, left, and right. Represent these points as two-dimensional points in the image coordinate system. Then, based on the camera calibration results, establish an affine mapping relationship from the image coordinate system to the physical measurement plane coordinate system. S3.2: Based on the current affine matrix, calculate the corresponding coordinates of the image coordinates of each contour point in the measurement plane. Then, perform line fitting on the physical coordinate contour points corresponding to each glass edge to obtain the geometric expression of the corresponding glass edge in physical space. S3.3: On each fitted straight line, based on the extreme position of the contour point in that direction, determine the two physical endpoints of the corresponding glass edge. Then, based on the two determined physical endpoints, calculate the straight-line distance of the corresponding glass edge in physical space to obtain the initial length value of each glass edge. Statistically calculate the obtained initial length values ​​of each glass edge to generate the corresponding initial glass size.

6. The method for measuring the size of liquid crystal substrate glass based on artificial intelligence and image analysis according to claim 5, characterized in that, The specific steps for adjusting the initial glass size using the intelligent prediction model described in step S5 are as follows: S4.1: Select multiple sets of original dimensions from the historical measurement process and match them one by one with the actual dimensions to construct a historical sample dataset. Then, use the sample dataset to train the model and establish a prediction model. S4.2: The initial side length data are used as input and sequentially imported into the size prediction model. The size prediction model performs multi-parameter mapping on the initial side length data, outputs the predicted side length size, compares the predicted side length size with the corresponding real size, and calculates the prediction error of each historical sample. S4.3: Calculate the prediction error of each historical sample to obtain the total loss error of the size prediction model, and minimize the total loss error of the size correction model to obtain the model parameters; S4.4: Input the initial dimensions obtained from online measurement into the optimized dimension prediction model to calculate the corrected final dimension result. Based on actual needs, decide whether to include this glass size in the historical sample set.

Images