Hyperbolic plate coating uniformity evaluation method and system based on machine vision

Abstract

The invention relates to the field of image processing, in particular to a hyperbolic plate coating uniformity evaluation method and system based on machine vision. Comprising the steps of obtaining a hyperbolic plate image; acquiring a curvature region image set according to the hyperbolic plate image; determining a visual angle image set according to the curvature area image set, and obtaining a brightness image set according to the visual angle image set; calculating the image quality of each image in the brightness image set, and determining a target visual angle image in each brightness image set; extracting an image corresponding to each target view angle image from the view angle image set to obtain a uniformity analysis image, and obtaining a hue analysis image according to the uniformity analysis image; calculating a gradient difference threshold value of the hue analysis image, and converting the hue analysis image according to the gradient difference threshold value to obtain a converted image; and calculating the uniformity of the uniformity analysis image according to the converted image, and determining the uniformity quality of the hyperbolic plate. According to the invention, the accuracy of uniformity detection on the coating of the hyperbolic plate can be improved.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more specifically to a machine vision-based method and system for evaluating the uniformity of hyperbolic plate coatings. Background Technology

[0002] Hyperbolic plates are complex three-dimensional curved surface plates with opposite curvatures (one positive and one negative) in two mutually perpendicular directions, resembling a saddle or cooling tower.

[0003] Existing methods often employ image analysis based on grayscale variance when evaluating coating uniformity.

[0004] However, due to the complex curvature changes on the surface of the hyperbolic plate, uneven illumination and projection distortion are easily caused, which greatly interferes with the calculation method that directly relies on the gray-scale variance, resulting in significant detection errors and making it difficult to achieve accurate evaluation. Summary of the Invention

[0005] This invention provides a machine vision-based method and system for evaluating the uniformity of hyperbolic plate coatings to solve existing problems.

[0006] The present invention provides a machine vision-based method for evaluating the uniformity of hyperboloid coatings, which employs the following technical solution: One embodiment of the present invention provides a machine vision-based method for evaluating the uniformity of hyperbolic coatings, the method comprising the following steps: Obtain the hyperbolic plate image, where the hyperbolic plate image is an RGB image; Obtain a curvature region image set from the hyperboloid image, wherein the curvature region image set includes images corresponding to different curvature regions; The viewpoint image set is determined based on the curvature region image set, and the viewpoint image set is converted into an HIS image and the I channel is extracted to obtain the brightness image set; Calculate the image quality of each image in the brightness image set and determine the target view image in each brightness image set, where the target view image is the image with the highest image quality in each view image set; The image corresponding to each target viewpoint image is extracted from the viewpoint image set to obtain the uniformity analysis image. The uniformity analysis image is then converted into an HIS image and the H channel is extracted to obtain the tone analysis image. Calculate the gradient difference threshold of the tone analysis image, and transform the tone analysis image according to the gradient difference threshold to obtain the transformed image; The uniformity of the hyperbolic plate is determined by calculating the uniformity of the converted image and analyzing the uniformity of the image.

[0007] Optionally, a set of curvature region images is obtained from the hyperbolic plate image, specifically including: Determine the 3D model of the triangular network based on the hyperbolic plate image; The curvature of each point in the 3D model of the triangular network is calculated based on the rate of change of the normal vector. Sort the curvature at each location point to obtain a curvature sequence; The curvature sequence is segmented using the Otsu multi-threshold segmentation method to obtain curvature segments; The first and second position points in the 3D model of the triangular network are merged to obtain the curvature region. The first and second position points are any position points in the 3D model of the triangular network. The first and second position points are adjacent to each other and are in the same curvature segment. The image corresponding to each curvature region is cropped from the hyperbolic plate image to obtain the curvature region image set.

[0008] Optionally, the viewpoint image set is determined based on the curvature region image set, specifically including: Each image of each curvature region in the curvature region image set is input into the trained neural radiation field model, and the trained neural radiation field model outputs the multi-view image corresponding to each image. The trained neural radiation field model is trained based on the hyperbolic plate image. Each image of each curvature region in the curvature region image set and its corresponding multi-view image are defined as the view image set.

[0009] Optionally, the image quality of each image in the brightness image set is calculated, specifically including: The Otsu multi-threshold segmentation method is used to detect each image in the brightness image set to obtain the segmentation threshold. The ratio of the number of pixels in the Nth image that are greater than the segmentation threshold of the Nth image to the total number of pixels in the Nth image is determined as the high-brightness ratio of the Nth image. Calculate the entropy value of the gray-level co-occurrence matrix of the Nth image in the brightness image set, and normalize the entropy value to obtain the gray-level entropy of the Nth image; The image quality of the Nth image is obtained by multiplying the proportion of high-brightness areas in the Nth image by the gray-level entropy of the Nth image.

[0010] Optionally, the gradient difference threshold of the tone analysis image is calculated, specifically including: Obtain the gradient of each pixel in the tone analysis image to obtain a gradient map; Pixels in the gradient map that are greater than a preset gradient threshold are identified as gradient pixels. The minimum gradient value among the gradient pixels is determined as the gradient threshold; Calculate the difference between the mode of the gradient in the gradient graph and the gradient threshold, take the absolute value of the difference, and calculate the ratio of the absolute value to the maximum gradient value in the gradient graph to obtain the gradient difference threshold.

[0011] Optionally, the tone analysis image is transformed according to a gradient difference threshold to obtain a transformed image, specifically including: The first tone pixel is determined from the tone analysis image, and the pixel that is adjacent to the first tone pixel in the eight-neighbor domain is determined as the second tone pixel. Calculate the difference between the hue value of the second hue pixel and the hue value of the first hue pixel, and divide the absolute value of the difference by the hue value of the first hue pixel to obtain the degree of anomalousness of the second hue pixel compared to the first hue pixel; When the anomaly level is greater than the gradient difference threshold and the number of second-tone pixels is greater than 3, the first-tone pixels are identified as difference pixels. The hue value of the second hue pixel corresponding to the maximum degree of anomaly is determined as the hue value of the difference pixel, and the difference pixel is redefined as the first hue pixel, until the degree of anomaly of the second hue pixel compared with the first hue pixel is less than the gradient difference threshold, or the degree of anomaly is greater than the gradient difference threshold and the number of second hue pixels is less than 4. Each pixel in the tone analysis image is converted as a first tone pixel to obtain the converted image.

[0012] Optionally, the uniformity of the image is analyzed by calculating the uniformity of the transformed image, specifically including: Calculate the tonal variance of all pixels in the transformed image to obtain the difference. The tonal layer is obtained by subtracting the tonal analysis image from the transformed image; The pixels with non-zero hue values ​​in the hue layer that correspond to the pixels in the transformed image are identified as transformed pixels, and the regions formed by the transformed pixels are identified as the transformed connected components. Get the sum of the number of transformations of pixels in the connected component; Normalize the sum of the number of conversions to obtain the conversion frequency; The product of the conversion frequency and the discrimination is determined as the uniformity of the uniformity analysis image, where the discrimination is obtained by taking the negative of the difference and using it as the exponent of the natural base.

[0013] Optionally, the uniformity quality of the hyperboloid is determined based on the uniformity of the uniformity analysis image, specifically including: The uniformity of the uniformity analysis image is determined as the uniformity of the corresponding image in the curvature region image set, thus obtaining the uniformity of each image in the curvature region image set. The coating uniformity of each curvature region in the hyperboloid image is determined based on the uniformity of each image in the curvature region image set. The overall uniformity of the hyperbolic plate is determined based on the coating uniformity in each curvature region of the hyperbolic plate image. When the overall uniformity is less than the preset first uniformity threshold, it is determined that the uniformity quality of the hyperbola has a significant defect; when the overall uniformity is greater than or equal to the preset first uniformity threshold, and the coating uniformity of the target curvature region is less than the preset second uniformity threshold, it is determined that the uniformity quality of the target curvature region of the hyperbola has a significant defect; when the overall uniformity is greater than or equal to the preset first uniformity threshold, and the coating uniformity is greater than or equal to the preset second uniformity threshold, it is determined that the uniformity quality of the hyperbola has no defects.

[0014] Optionally, the 3D model of the triangular network is determined based on the hyperbolic plate image, specifically including: The hyperbolic plate image and the viewpoint parameters corresponding to each hyperbolic plate image are input into the neural radiation field model for training, and the trained neural radiation field model is obtained. Three-dimensional density field data is constructed by using a trained neural radiation field model. The three-dimensional density field data is then input into the moving cube algorithm model, which outputs a triangular network three-dimensional model.

[0015] This invention proposes a machine vision-based hyperboloid coating uniformity evaluation system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the machine vision-based hyperboloid coating uniformity evaluation method described above.

[0016] The beneficial effects of the technical solution of the present invention are: In this embodiment of the invention, the image obtained by the hyperboloid plate is divided into partitions by curvature, thereby obtaining the image of each partition that is least affected by illumination. The image is then transformed to obtain a transformed image with magnified differences. The coating uniformity evaluation value of each image is obtained based on the degree of difference and the difficulty of transformation, thereby improving the accuracy of uniformity detection of the coating of the hyperboloid plate. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a machine vision-based method for evaluating the uniformity of hyperbolic plate coatings, as provided in one embodiment of the present invention. Figure 2This is a schematic diagram showing the corresponding images from multiple perspectives; Figure 3 This is a structural diagram of a machine vision-based hyperbolic plate coating uniformity evaluation system provided in one embodiment of the present invention. Detailed Implementation

[0019] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0021] The following description, in conjunction with the accompanying drawings, details a specific scheme for a machine vision-based method for evaluating the uniformity of hyperbolic plate coatings provided by this invention.

[0022] This invention provides a machine vision-based method and system for evaluating the uniformity of hyperbolic plate coatings. Please refer to [link to relevant documentation]. Figure 1 The diagram illustrates a flowchart of a machine vision-based method for evaluating the uniformity of a hyperbolic plate coating, according to an embodiment of the present invention. The method includes the following steps: S101. Obtain the hyperbolic plate image, wherein the hyperbolic plate image is an RGB image.

[0023] In one specific embodiment, the hardware and parameters for effectively acquiring hyperbolic plate images are as follows: Hardware includes: Camera: High-resolution, global shutter industrial camera to ensure no motion blur. Lens: Preferred image-side telecentric lens to fundamentally eliminate perspective distortion. Illumination: Integrating sphere dome lighting must be used to provide uniform diffused light, completely eliminating reflections and shadows. Rotary stage: High-precision motorized rotary stage with positioning accuracy at the arcsecond level.

[0024] The parameters include: System calibration: High-precision camera calibration is performed using a calibration board before acquisition to obtain lens distortion parameters. Path planning: The rotation angle interval (e.g., 2°) is calculated based on the camera's field of view to ensure that the overlap rate of adjacent images is >60% to meet the requirements of 3D reconstruction. Synchronous acquisition: After the rotary table reaches the predetermined angle, a hardware trigger signal is sent to the camera for exposure to ensure that the angle and image strictly correspond. Parameter settings: Save using a lossless format (e.g., TIFF); the exposure time is set to avoid overexposure; the aperture is stopped down to the optimal sharpness range.

[0025] S102. Obtain a curvature region image set based on the hyperbolic plate image, wherein the curvature region image set includes images corresponding to different curvature regions.

[0026] In this embodiment, obtaining a curvature region image set based on the hyperboloid image specifically includes: Determine the 3D model of the triangular network based on the hyperbolic plate image; The curvature of each point in the 3D model of the triangular network is calculated based on the rate of change of the normal vector. Sort the curvature at each location point to obtain a curvature sequence; The curvature sequence is segmented using the Otsu multi-threshold segmentation method to obtain curvature segments; The first and second position points in the 3D model of the triangular network are merged to obtain the curvature region. The first and second position points are any position points in the 3D model of the triangular network. The first and second position points are adjacent to each other and are in the same curvature segment. The image corresponding to each curvature region is cropped from the hyperbolic plate image to obtain the curvature region image set.

[0027] Optionally, the 3D model of the triangular network is determined based on the hyperbolic plate image, specifically including: The hyperbolic plate image and the viewpoint parameters corresponding to each hyperbolic plate image are input into the neural radiation field model for training, and the trained neural radiation field model is obtained. Three-dimensional density field data is constructed by using a trained neural radiation field model. The three-dimensional density field data is then input into the moving cube algorithm model, which outputs a triangular network three-dimensional model.

[0028] For example, a triangular network 3D model is used to represent the specific 3D structure of a hyperbolic plate surface, which consists of countless vertices and triangular facets.

[0029] Due to the irregular curved surface structure of the hyperbolic plate, some areas will always reflect light when the entire hyperbolic plate is photographed. Therefore, it is impossible to directly evaluate the uniformity of the coating by factors such as the brightness and hue of different pixels. Therefore, in this embodiment, the hyperbolic plate is divided into sections according to the curvature. The lighting conditions of the same section are similar, so uniformity can be calculated.

[0030] For each point on the 3D model, the curvature of each location in the model is obtained based on the rate of change of the normal vector.

[0031] All curvature values ​​are arranged in ascending order to obtain an ascending sequence (in this embodiment, all curvatures are sorted in ascending order; in practice, they can also be sorted in descending order, and no specific restriction is made here). The ascending sequence is then processed using the Otsu multi-threshold segmentation method to obtain multiple curvature segments. Curvature values ​​within the same curvature segment are similar, while curvature values ​​in different curvature segments differ significantly.

[0032] In the 3D model, for each point, the curvature value of each of its eight neighboring points is compared with that point. If the curvature values ​​belong to the same segment, the two points are merged. Through multiple mergings, multiple curvature regions can be obtained, and the curvature values ​​of the same curvature region belong to the same segment. For example, a hyperboloid can be divided into three curvature regions: upper, middle, and lower. Within each curvature region, the influence of lighting on the pixels within that region is similar, thus allowing for uniformity calculations.

[0033] For each curvature region, the corresponding region is cropped from the captured image to obtain multiple images of that curvature region, which are denoted as the curvature region image set.

[0034] S103. Determine the viewpoint image set based on the curvature region image set, convert the viewpoint image set into an HIS image and extract the I channel to obtain the brightness image set.

[0035] In this embodiment, determining the viewpoint image set based on the curvature region image set specifically includes: Each image of each curvature region in the curvature region image set is input into the trained neural radiation field model, and the trained neural radiation field model outputs the multi-view image corresponding to each image. The trained neural radiation field model is trained based on the hyperbolic plate image. Each image of each curvature region in the curvature region image set and its corresponding multi-view image are defined as the view image set.

[0036] For example, on each curvature region image set, images of the curvature region at different rotation angles can be obtained, and adjacent images captured have overlapping areas (based on the shooting process in S101, here different rotation angles correspond to different images in the hyperbolic plate image, and by taking an image at one rotation angle, the hyperbolic plate image corresponding to each rotation angle can be obtained).

[0037] like Figure 2 As shown, Figure 2 This is a schematic diagram showing the correspondence between multi-view images. Figure 2In this method, by using a camera with a fixed viewpoint and rotating a hyperbolic plate, images can be obtained at different rotation angles. Then, through neural radiation fields, images from multiple other viewpoints can be obtained at each rotation angle. The figure only shows four images from the other viewpoints at each rotation angle (for example only), but there may actually be more than four images from other viewpoints.

[0038] The image at the orthogonal viewpoint corresponding to each rotation angle and the images at other viewpoints are denoted as the viewpoint image set for that rotation angle.

[0039] S104. Calculate the image quality of each image in the brightness image set and determine the target view image in each brightness image set, wherein the target view image is the image with the highest image quality in each view image set.

[0040] In this embodiment, calculating the image quality of each image in the brightness image set specifically includes: The Otsu multi-threshold segmentation method is used to detect each image in the brightness image set to obtain the segmentation threshold. The ratio of the number of pixels in the Nth image that are greater than the segmentation threshold of the Nth image to the total number of pixels in the Nth image is determined as the high-brightness ratio of the Nth image. Calculate the entropy value of the gray-level co-occurrence matrix of the Nth image in the brightness image set, and normalize the entropy value to obtain the gray-level entropy of the Nth image; The image quality of the Nth image is obtained by multiplying the proportion of high-brightness areas in the Nth image by the grayscale entropy of the Nth image.

[0041] For example, for each image in the viewpoint image set at each rotation angle, the RGB image is converted to an HSI image, and then the I channel is extracted to obtain the luminance image. The luminance image is segmented using the Otsu thresholding method to obtain a threshold k. Pixels greater than the threshold k are designated as highlights, and the ratio of the number of highlights to the total number of pixels in the image (i.e., the highlight ratio) is represented by b. For each image in the viewpoint image set at each rotation angle, the entropy value of the gray-level co-occurrence matrix is ​​calculated and normalized to obtain the gray-level entropy s. The larger s is, the richer the texture details and the more diffuse reflection in the image.

[0042] For each image in the viewpoint image set for each rotation angle, calculate the product of b and s as the image quality, and determine the image with the highest image quality as the target viewpoint image for each rotation angle.

[0043] S105. Extract the image corresponding to each target viewpoint image from the viewpoint image set to obtain a uniformity analysis image. Convert the uniformity analysis image into an HIS image and extract the H channel to obtain a tone analysis image.

[0044] S106. Calculate the gradient difference threshold of the tone analysis image, and transform the tone analysis image according to the gradient difference threshold to obtain the transformed image.

[0045] In this embodiment, calculating the gradient difference threshold of the tone analysis image specifically includes: Obtain the gradient of each pixel in the tone analysis image to obtain a gradient map; Pixels in the gradient map that are greater than a preset gradient threshold are identified as gradient pixels. The minimum gradient value among the gradient pixels is determined as the gradient threshold; Calculate the difference between the mode of the gradient in the gradient graph and the gradient threshold, take the absolute value of the difference, and calculate the ratio of the absolute value to the maximum gradient value in the gradient graph to obtain the gradient difference threshold.

[0046] The tonal analysis image is transformed based on a gradient difference threshold to obtain the transformed image, specifically including: The first tone pixel is determined from the tone analysis image, and the pixel that is adjacent to the first tone pixel in the eight-neighbor domain is determined as the second tone pixel. Calculate the difference between the hue value of the second hue pixel and the hue value of the first hue pixel, and divide the absolute value of the difference by the hue value of the first hue pixel to obtain the degree of anomalousness of the second hue pixel compared to the first hue pixel; When the anomaly level is greater than the gradient difference threshold and the number of second-tone pixels is greater than 3, the first-tone pixels are identified as difference pixels. The hue value of the second hue pixel corresponding to the maximum degree of anomaly is determined as the hue value of the difference pixel, and the difference pixel is redefined as the first hue pixel, until the degree of anomaly of the second hue pixel compared with the first hue pixel is less than the gradient difference threshold, or the degree of anomaly is greater than the gradient difference threshold and the number of second hue pixels is less than 4. Each pixel in the tone analysis image is converted as a first tone pixel to obtain the converted image.

[0047] For example, if the coating is uniform, the hue values ​​of different pixels in the uniformity analysis image are similar. For each rotation angle, the corresponding uniformity analysis image is converted to the HSI space, and the data of the H channel, i.e., the hue channel data, is extracted to obtain the hue analysis image at each rotation angle.

[0048] To address the problem that directly calculating hue variance makes it difficult to effectively identify subtle differences, this paper proposes to amplify hue differences and generate transformed images. By combining the significance of these differences with the difficulty of the transformation, a quantitative assessment of coating uniformity can be achieved.

[0049] For the tone analysis image of each curvature region, if the difference between the tone values ​​of pixels in the converted image is greater and the conversion difficulty is smaller, it indicates that the original image itself has unstable and easily amplified non-uniform factors, that is, the non-uniformity of the hyperbolic plate coating corresponding to the original image is greater.

[0050] For each tone analysis image, a corresponding gradient map is obtained. The pixel value of each pixel in the gradient map represents the gradient value of that pixel. Pixels in the gradient map that are greater than a preset gradient threshold are identified as gradient pixels.

[0051] In a specific embodiment, the preset gradient threshold can be obtained by sorting the gradient values ​​in descending order and determining the smallest gradient value among the top 5% of gradient values ​​as the preset gradient threshold. The preset gradient threshold can also be set according to the actual situation or determined based on historical data, and no specific restrictions are imposed here.

[0052] The minimum gradient value among the gradient pixels is denoted as the gradient threshold; the absolute value of the difference between the gradient threshold and the mode of the gradient in the gradient map is calculated, and the ratio of the absolute value of the difference to the maximum gradient value is calculated. This ratio is used as the gradient difference threshold y.

[0053] For any pixel in each tone analysis image, the ratio of the absolute value of the difference between the tone value of this pixel and the tone value of each of its eight neighboring pixels to the tone value of this pixel is calculated. This ratio is recorded as the degree of anomalousness of the neighboring pixels compared to this pixel. For any pixel in each tone analysis image, if more than half of the pixels in its eight neighboring pixels have a tone value greater than y that differs from this pixel's tone value, then this pixel is recorded as a difference pixel. Simultaneously, the tone value of this difference pixel is converted to the tone value of the pixel in its neighboring pixels with the highest degree of anomalousness compared to this pixel. This conversion is repeated multiple times until no pixels in the neighborhood meet the condition, at which point the conversion stops. After all pixels have been converted, the converted image is obtained. In this image, any pixel in the tone analysis image can be a first tone pixel, and any pixel in the eight neighboring pixels of a first tone pixel can be a second tone pixel.

[0054] S107. Calculate the uniformity of the uniformity analysis image based on the transformed image, and determine the uniformity quality of the hyperboloid based on the uniformity of the uniformity analysis image.

[0055] In this embodiment, the uniformity of the image is analyzed by calculating the uniformity of the transformed image, specifically including: Calculate the tonal variance of all pixels in the transformed image to obtain the difference. The tonal layer is obtained by subtracting the tonal analysis image from the transformed image; The pixels with non-zero hue values ​​in the hue layer that correspond to the pixels in the transformed image are identified as transformed pixels, and the regions formed by the transformed pixels are identified as the transformed connected components. Get the sum of the number of transformations of pixels in the connected component; Normalize the sum of the number of conversions to obtain the conversion frequency; The product of the conversion frequency and the discrimination is determined as the uniformity of the uniformity analysis image, where the discrimination is obtained by taking the negative of the difference and using it as the exponent of the natural base.

[0056] The uniformity quality of the hyperboloid plate is determined based on the uniformity of the uniformity analysis image, specifically including: The uniformity of the uniformity analysis image is determined as the uniformity of the corresponding image in the curvature region image set, thus obtaining the uniformity of each image in the curvature region image set. The coating uniformity of each curvature region in the hyperboloid image is determined based on the uniformity of each image in the curvature region image set. The overall uniformity of the hyperbolic plate is determined based on the coating uniformity in each curvature region of the hyperbolic plate image. When the overall uniformity is less than the preset first uniformity threshold, it is determined that the uniformity quality of the hyperbola has a significant defect; when the overall uniformity is greater than or equal to the preset first uniformity threshold, and the coating uniformity of the target curvature region is less than the preset second uniformity threshold, it is determined that the uniformity quality of the target curvature region of the hyperbola has a significant defect; when the overall uniformity is greater than or equal to the preset first uniformity threshold, and the coating uniformity is greater than or equal to the preset second uniformity threshold, it is determined that the uniformity quality of the hyperbola has no defects.

[0057] For example, the tonal variance of all pixels in the transformed image is calculated as the difference c of pixel values ​​in the transformed image. The smaller the value, the better the uniformity of the corresponding coating on the original image.

[0058] The tonal analysis image is subtracted from the transformed image to obtain a tonal layer. The pixels in the transformed image corresponding to the pixels with tonal values ​​not equal to 0 in the tonal layer are identified as transformed pixels. The regions formed by the transformed pixels are identified as transformed connected regions. Alternatively, the following steps can be taken: the pixels in the transformed image that have obtained new tonal values ​​through transformation are recorded as transformed pixels, and the connected regions enclosed by the transformed pixels are recorded as transformed connected regions.

[0059] For each transformed connected component, the number of transformations required for each pixel in that component to become a transformed pixel is determined. For example, a pixel's eight neighbors might not initially meet the transformation criteria. It is only after another pixel undergoes multiple transformations that the eight neighbors of the transformed pixel meet the criteria, thus making the transformed pixel a transformed pixel. For each transformed pixel, the more transformations, the greater the transformation difficulty, and consequently, the better the uniformity of the original image.

[0060] The number of transformations for each transformed pixel is counted, and then the sum of all transformation counts is obtained. This sum is then normalized to obtain the transformation frequency *g*. For each transformed image, the transformation frequency *g* is calculated. The result is used as the uniformity of the corresponding original image, where the formula for calculating the discrimination is: The uniformity of the hyperbolic plate region corresponding to each rotated image of each curvature region can be obtained through calculation.

[0061] For each curvature region, the uniformity of the hyperboloid region corresponding to each rotated image can be obtained. The average of all uniformities is then used as the coating uniformity for that curvature region. For all curvature regions, the average coating uniformity of all curvature regions is calculated as the overall uniformity of the hyperboloid.

[0062] The above process yielded coating uniformity in regions of different curvatures, i.e., coating uniformity in local areas and uniformity in the overall coating area. Then, judgments can be made based on their respective uniformity thresholds (these thresholds can be the manufacturer's testing thresholds, which can be set by the manufacturer using historical data, without specific numerical limitations). If the overall uniformity meets the preset first uniformity threshold, then check whether the uniformity of each curvature region meets the preset second uniformity threshold. If it does not meet the preset second uniformity threshold, repaint or perform other operations on different curvature regions to make the uniformity meet the conditions. If the overall uniformity meets the preset first uniformity threshold and each curvature region also meets the preset second uniformity threshold, then it proves that the hyperbolic board meets the factory conditions.

[0063] Optionally, the preset first uniformity threshold and the preset second uniformity threshold can be set and adjusted in real time according to the actual situation. There is no specific numerical limit on the preset first uniformity threshold and the preset second uniformity threshold. In a preferred embodiment, the preset first uniformity threshold and the preset second uniformity threshold can be 0.7 and 0.8, respectively.

[0064] This invention is now complete.

[0065] In summary, in this embodiment of the invention, the image obtained by the hyperbolic plate is partitioned by curvature to obtain the image of each partition that is least affected by illumination. The image is then transformed to obtain a transformed image with magnified differences. The coating uniformity evaluation value of each image is obtained based on the degree of difference and the difficulty of transformation, thereby improving the accuracy of uniformity detection of the soil layer of the hyperbolic plate.

[0066] This invention also proposes a machine vision-based system for evaluating the uniformity of hyperbolic plate coatings; please refer to [link to relevant documentation]. Figure 3 The diagram shows a structural diagram of a machine vision-based hyperbolic plate coating uniformity evaluation system provided in an embodiment of the present invention. The system includes: a data acquisition module 101, a data processing module 102, and a quality evaluation module 103.

[0067] The data acquisition module 101 is used to acquire the hyperbolic plate image, wherein the hyperbolic plate image is an RGB image; Data processing module 102 is used to obtain a curvature region image set based on the hyperbolic plate image, wherein the curvature region image set includes images corresponding to different curvature regions; determine a viewpoint image set based on the curvature region image set, convert the viewpoint image set into an HIS image and extract the I channel to obtain a luminance image set; calculate the image quality of each image in the luminance image set and determine the target viewpoint image in each luminance image set, wherein the target viewpoint image is the image with the highest image quality in each viewpoint image set; extract the image corresponding to each target viewpoint image from the viewpoint image set to obtain a uniformity analysis image, convert the uniformity analysis image into an HIS image and extract the H channel to obtain a tone analysis image; calculate the gradient difference threshold of the tone analysis image, and convert the tone analysis image according to the gradient difference threshold to obtain a converted image; The quality evaluation module 103 is used to calculate the uniformity of the uniformity analysis image based on the transformed image, and to determine the uniformity quality of the hyperbolic plate based on the uniformity of the uniformity analysis image.

[0068] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the machine vision-based hyperbolic plate coating uniformity evaluation system and the machine vision-based hyperbolic plate coating uniformity evaluation method embodiment provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiment, which will not be repeated here.

[0069] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0070] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based method for evaluating the uniformity of coatings on hyperboloidal plates, characterized in that, include: Obtain the hyperbolic plate image, where the hyperbolic plate image is an RGB image; Obtain a curvature region image set from the hyperboloid image, wherein the curvature region image set includes images corresponding to different curvature regions; The viewpoint image set is determined based on the curvature region image set, and the viewpoint image set is converted into an HIS image and the I channel is extracted to obtain the brightness image set; Calculate the image quality of each image in the brightness image set and determine the target view image in each brightness image set, where the target view image is the image with the highest image quality in each view image set; The image corresponding to each target viewpoint image is extracted from the viewpoint image set to obtain the uniformity analysis image. The uniformity analysis image is then converted into an HIS image and the H channel is extracted to obtain the tone analysis image. Calculate the gradient difference threshold of the tone analysis image, and transform the tone analysis image according to the gradient difference threshold to obtain the transformed image; The uniformity of the hyperbolic plate is determined by calculating the uniformity of the converted image and analyzing the uniformity of the image.

2. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The step of obtaining the curvature region image set based on the hyperbolic plate image specifically includes: Determine the 3D model of the triangular network based on the hyperbolic plate image; The curvature of each point in the 3D model of the triangular network is calculated based on the rate of change of the normal vector. Sort the curvature at each location point to obtain a curvature sequence; The curvature sequence is segmented using the Otsu multi-threshold segmentation method to obtain curvature segments; The first and second position points in the 3D model of the triangular network are merged to obtain the curvature region. The first and second position points are any position points in the 3D model of the triangular network. The first and second position points are adjacent to each other and are in the same curvature segment. The image corresponding to each curvature region is cropped from the hyperbolic plate image to obtain the curvature region image set.

3. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The determination of the viewpoint image set based on the curvature region image set specifically includes: Each image of each curvature region in the curvature region image set is input into the trained neural radiation field model, and the trained neural radiation field model outputs the multi-view image corresponding to each image. The trained neural radiation field model is trained based on the hyperbolic plate image. Each image of each curvature region in the curvature region image set and its corresponding multi-view image are defined as the view image set.

4. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The image quality of each image in the brightness calculation image set specifically includes: The Otsu multi-threshold segmentation method is used to detect each image in the brightness image set to obtain the segmentation threshold. The ratio of the number of pixels in the Nth image that are greater than the segmentation threshold of the Nth image to the total number of pixels in the Nth image is determined as the high-brightness ratio of the Nth image. Calculate the entropy value of the gray-level co-occurrence matrix of the Nth image in the brightness image set, and normalize the entropy value to obtain the gray-level entropy of the Nth image; The image quality of the Nth image is obtained by multiplying the proportion of high-brightness areas in the Nth image by the grayscale entropy of the Nth image.

5. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The calculation of the gradient difference threshold of the tone analysis image specifically includes: Obtain the gradient of each pixel in the tone analysis image to obtain a gradient map; Pixels in the gradient map that are greater than a preset gradient threshold are identified as gradient pixels. The minimum gradient value among the gradient pixels is determined as the gradient threshold; Calculate the difference between the mode of the gradient in the gradient graph and the gradient threshold, take the absolute value of the difference, and calculate the ratio of the absolute value to the maximum gradient value in the gradient graph to obtain the gradient difference threshold.

6. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The process of transforming the tone analysis image based on a gradient difference threshold to obtain a transformed image specifically includes: The first tone pixel is determined from the tone analysis image, and the pixel that is adjacent to the first tone pixel in the eight-neighbor domain is determined as the second tone pixel. Calculate the difference between the hue value of the second hue pixel and the hue value of the first hue pixel, and divide the absolute value of the difference by the hue value of the first hue pixel to obtain the degree of anomalousness of the second hue pixel compared to the first hue pixel; When the anomaly level is greater than the gradient difference threshold and the number of second-tone pixels is greater than 3, the first-tone pixels are identified as difference pixels. The hue value of the second hue pixel corresponding to the maximum degree of anomaly is determined as the hue value of the difference pixel, and the difference pixel is redefined as the first hue pixel, until the degree of anomaly of the second hue pixel compared with the first hue pixel is less than the gradient difference threshold, or the degree of anomaly is greater than the gradient difference threshold and the number of second hue pixels is less than 4. Each pixel in the tone analysis image is converted as a first tone pixel to obtain the converted image.

7. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The process of calculating uniformity analysis of the transformed image specifically includes: Calculate the tonal variance of all pixels in the transformed image to obtain the difference. The tonal layer is obtained by subtracting the tonal analysis image from the transformed image; The pixels with non-zero hue values ​​in the hue layer that correspond to the pixels in the transformed image are identified as transformed pixels, and the regions formed by the transformed pixels are identified as the transformed connected components. Get the sum of the number of transformations of pixels in the connected component; Normalize the sum of the number of conversions to obtain the conversion frequency; The product of the conversion frequency and the discrimination is determined as the uniformity of the uniformity analysis image, where the discrimination is obtained by taking the negative of the difference and using it as the exponent of the natural base.

8. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 1, characterized in that, The determination of the uniformity quality of the hyperboloid based on the uniformity of the uniformity analysis image specifically includes: The uniformity of the uniformity analysis image is determined as the uniformity of the corresponding image in the curvature region image set, thus obtaining the uniformity of each image in the curvature region image set. The coating uniformity of each curvature region in the hyperboloid image is determined based on the uniformity of each image in the curvature region image set. The overall uniformity of the hyperbolic plate is determined based on the coating uniformity in each curvature region of the hyperbolic plate image. When the overall uniformity is less than the preset first uniformity threshold, it is determined that the uniformity quality of the hyperbola has a significant defect; when the overall uniformity is greater than or equal to the preset first uniformity threshold, and the coating uniformity of the target curvature region is less than the preset second uniformity threshold, it is determined that the uniformity quality of the target curvature region of the hyperbola has a significant defect; when the overall uniformity is greater than or equal to the preset first uniformity threshold, and the coating uniformity is greater than or equal to the preset second uniformity threshold, it is determined that the uniformity quality of the hyperbola has no defects.

9. The machine vision-based method for evaluating the uniformity of hyperbolic plate coatings according to claim 2, characterized in that, The process of determining the 3D model of the triangular network based on the hyperbolic plate image specifically includes: The hyperbolic plate image and the viewpoint parameters corresponding to each hyperbolic plate image are input into the neural radiation field model for training, and the trained neural radiation field model is obtained. Three-dimensional density field data is constructed by using a trained neural radiation field model. The three-dimensional density field data is then input into the moving cube algorithm model, which outputs a triangular network three-dimensional model.

10. A machine vision-based hyperboloid coating uniformity evaluation system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the machine vision-based hyperbolic plate coating uniformity evaluation method as described in any one of claims 1-9.

Images