Composite board ceramic coating quality detection method and system

Abstract

The invention belongs to the technical field of coating quality detection, and particularly relates to a composite board ceramic coating quality detection method and system, and the method comprises the steps: obtaining the local neighborhood information of each pixel point according to a sliding window, calculating the gray local contrast, and constructing a gray suppression factor; taking the normalized average value of the gradient amplitudes of the pixels in the sliding window as a texture feature index; obtaining a background consistency enhancement coefficient based on the gray suppression factors of the normal candidate set and the abnormal candidate set and the standard deviation of all the gray suppression factors; obtaining a defect texture protection coefficient by combining the texture feature indexes of the missing coating and reflection candidate set and the standard deviation of all the texture feature indexes, and constructing the sparse penalty weight of each pixel point; weighting the regularization parameters to obtain dynamic regularization parameters, and constructing and solving an objective function to obtain a sparse abnormal matrix; and separating a missed coating area from the sparse abnormal matrix in combination with a texture feature index. According to the invention, the accuracy of quality detection of the ceramic coating of the composite board is improved.

Description

Technical Field

[0001] This invention relates to the field of coating quality testing technology, and in particular to a method and system for testing the quality of ceramic coatings on composite panels. Background Technology

[0002] In the composite panel manufacturing industry, spraying a ceramic coating onto the surface of a metal substrate is a common process. This process aims to combine the toughness of metal with the wear resistance and corrosion resistance of ceramics to produce high-quality composite panels. However, if the ceramic coating is not applied properly during production, exposing the metal substrate, it is a very serious quality defect. This can easily lead to quality problems such as scratches and corrosion in the composite panel within a time that is far shorter than its design life. In severe cases, it can even cause safety accidents.

[0003] Since slight vibrations in the metal substrate can cause strong light spots in the uncoated areas, interfering with the accuracy of detecting the quality of the ceramic coating on the composite board surface, existing technologies employ robust principal component analysis (RPCA) to decompose the surface image of the composite board into a low-rank background matrix representing the normal coating and a sparse foreground matrix representing coating defects, in order to achieve accurate separation.

[0004] However, traditional RPCA algorithms typically employ globally fixed regularization parameters. To balance the weights of low-rank and sparse terms, the data characteristics of strong reflective noise with high brightness and large area on the metal substrate surface differ greatly from those of missing coating defects with low brightness and small area. If a large regularization parameter is set to suppress strong reflection, the algorithm will filter out tiny real missing coating defects as background, leading to an increase in the false negative rate. If a small regularization parameter is set to retain tiny details, the algorithm will misjudge metal reflective spots as sparse defects, leading to an increase in the false positive rate. Both of these will reduce the accuracy of the quality inspection of ceramic coating on the surface of composite panels. Summary of the Invention

[0005] To address the technical problem that traditional RPCA algorithms, due to their use of fixed regularization parameters, struggle to achieve a balance between suppressing strong reflective noise and preserving minute real defects, leading to increased false negative or false positive rates and ultimately affecting the accuracy of composite plate ceramic coating quality inspection, this invention provides a method for inspecting the quality of composite plate ceramic coatings.

[0006] In a first aspect, the present invention provides a method for quality inspection of ceramic coatings on composite panels, comprising: obtaining a grayscale analysis image; defining a sliding window on the grayscale analysis image and extracting local neighborhood information of each pixel; obtaining the local grayscale contrast of each pixel based on the local neighborhood information and the grayscale value of the pixel; constructing a grayscale suppression factor for each pixel according to the distribution of the local grayscale contrast of all pixels; using the normalized result of the average gradient magnitude of all pixels within the sliding window as a texture feature index for each pixel; and based on the difference in grayscale suppression factors between the normal candidate set and the abnormal candidate set, and the standard deviation of the grayscale suppression factors of all pixels. The background consistency enhancement coefficient is obtained; based on the difference in texture feature indices between the candidate sets of missing coatings and the candidate sets of reflections, and the standard deviation of texture feature indices of all pixels, a defect texture protection coefficient is obtained; and combined with the background consistency enhancement coefficient, grayscale suppression factor, and texture feature indices, a sparse penalty weight for each pixel is constructed; the sparse penalty weight is used to weight a globally fixed regularization parameter to obtain a dynamic regularization parameter; an optimized objective function is constructed based on the dynamic regularization parameter, and the objective function is solved to obtain a sparse anomaly matrix; the missing coating area is separated from the sparse anomaly matrix by combining texture feature indices, and is used as the quality inspection result of the composite plate ceramic coating and output.

[0007] This invention extracts local neighborhood information of pixels through a sliding window, constructs a gray-level suppression factor by combining local gray-level contrast, and constructs texture feature indicators based on the gradient magnitude of the Sobel operator. This accurately distinguishes normal coatings, reflective areas, and areas with missing coatings, enhancing the discriminative power. By constructing differentiated sparse penalty weights, the fixed regularization parameters of robust principal component analysis are optimized into dynamic regularization parameters, avoiding the problem of traditional algorithms struggling to balance reflective suppression and missing coating retention. Solving the optimized objective function can accurately separate normal coatings from abnormal areas, and combining texture feature indicators further filters out areas with missing coatings, significantly reducing the false negative and false positive rates, improving detection accuracy and reliability, adapting to the high-precision detection needs of industry, and ensuring the quality and safety of composite board products.

[0008] Preferably, the method for obtaining the local grayscale contrast of each pixel includes: ;in, It is a pixel. Local grayscale contrast; It is a pixel. grayscale value; It is the surrounding pixels grayscale value; It is the index of the pixel located at the center of the sliding window; It is the index of the peripheral pixels located outside the center of the sliding window; yes A sliding window of varying sizes, in which It is an odd number greater than 1; It is a pixel. With surrounding pixels The Euclidean distance between them; Based on pixels Centered sliding window Inside, pixels The standard deviation of the Euclidean distance from all surrounding pixels; It is a sliding window Contains pixels The total number of all pixels, including those in the array; It is a natural exponential function; It is an absolute value function.

[0009] This method integrates pixel grayscale differences with spatial location correlations, weighting grayscale differences within a sliding window using an exponential decay function of Euclidean distance. This highlights the dominant influence of neighboring pixels on the contrast of the central pixel and adaptively normalizes the distance weights using the standard deviation of distances within the window, effectively adapting to the spatial distribution characteristics of different windows. Simultaneously, the calculation results are normalized using the total number of pixels in the window, ensuring the consistency and comparability of contrast values. This method considers both the grayscale variation characteristics of local regions and the spatial distribution patterns of pixels, avoiding the contrast distortion problems caused by traditional methods that rely solely on grayscale differences or fixed weights. It improves the accuracy and robustness of local contrast representation, and its computational logic is simple and efficient, making it suitable for the accurate extraction of local detail features in various computer vision and image processing scenarios.

[0010] Preferably, the step of constructing a grayscale suppression factor for each pixel based on the distribution of local grayscale contrast of all pixels includes: ;in, It is a pixel. Gray-scale inhibition factor; It is a pixel. Local grayscale contrast; It is the average local grayscale contrast of all pixels; It is the standard deviation of the local grayscale contrast of all pixels; It is a natural exponential function.

[0011] This method extracts the global mean and global standard deviation of the local grayscale contrast of all pixels, and constructs a grayscale suppression factor using an exponential function of a normal distribution. This allows the suppression intensity to adaptively match the global contrast distribution: for pixels with local grayscale contrast close to the global mean, the suppression factor is close to 1, resulting in weak grayscale suppression to preserve the integrity of the region; for pixels with local grayscale contrast significantly deviating from the global mean, the grayscale suppression factor decreases exponentially, achieving targeted and strong suppression. This design avoids the problems of poor adaptability, over-suppression, or under-suppression associated with traditional fixed thresholds or locally independent calculation of suppression factors. By establishing a suppression benchmark through global statistics, it accurately distinguishes key details from redundant information, providing a precise basis for weight adjustment in subsequent image processing, and improving the robustness and rationality of the overall processing effect.

[0012] Preferably, the method for obtaining the normal candidate set and the abnormal candidate set includes: calculating the average and standard deviation of the grayscale suppression factor of all pixels, denoted as , respectively. The normal threshold is set according to the material characteristics of the ceramic coating of the composite panel and the required testing accuracy. The normal threshold The range of values ​​is set to to Pixels with a grayscale suppression factor greater than or equal to the normal threshold are classified as normal candidate sets belonging to the normal coating region; pixels with a grayscale suppression factor less than the normal threshold are classified as abnormal candidate sets belonging to the abnormal region.

[0013] Preferably, obtaining the background consistency enhancement coefficient based on the difference in grayscale suppression factors between the normal candidate set and the abnormal candidate set, and the standard deviation of the grayscale suppression factors of all pixels, includes: ;in, It is the background consistency enhancement coefficient; It is the average value of all grayscale suppression factors in the normal candidate set; It is the average value of all grayscale suppression factors in the abnormal candidate set; It is the standard deviation of the grayscale suppression factor for all pixels.

[0014] This method calculates the difference in mean grayscale suppression factors between normal and abnormal candidate sets, and combines this with the global standard deviation of the grayscale suppression factors of all pixels to obtain a background consistency enhancement coefficient. This coefficient accurately quantifies the degree of difference in grayscale suppression characteristics between the two sets, while effectively offsetting the interference of global distribution fluctuations on the difference representation. This design avoids the problems of inaccurate quantization and poor adaptability caused by traditional single feature or fixed parameter calculations, and achieves adaptive adjustment of the enhancement coefficient. It provides a reliable strength basis for subsequent background optimization, target differentiation, and other processing, improving the accuracy and stability of overall image processing.

[0015] Preferably, the method for obtaining the candidate set of missing coating and the candidate set of reflection includes: obtaining the standard deviation of the texture feature index of all pixels, denoted as... Next, for the anomaly candidate set belonging to the abnormal region, the texture feature index of each pixel in the anomaly candidate set is obtained, and the distribution of the texture feature index is used to calculate the anomaly threshold. Texture feature indicators greater than or equal to the anomaly threshold Pixels are divided into a candidate set of missing areas; texture feature indicators less than an abnormal threshold are considered. The pixels are divided into a set of reflective candidate points belonging to the reflective area.

[0016] Preferably, obtaining the defect texture protection coefficient based on the difference in texture feature indices between the missing coating candidate set and the reflection candidate set, and the standard deviation of the texture feature indices of all pixels, includes: ;in, It is the defect texture protection coefficient; It is the average of the texture feature indices of all pixels in the candidate set of missing paint. It is the average of the texture feature indices of all pixels in the reflective candidate set; It is the standard deviation of the texture feature index of all pixels.

[0017] This method calculates the difference in mean texture feature indices between the candidate sets of missing coatings and those of reflective defects, and combines this with the global standard deviation of texture feature indices for all pixels to obtain a defect texture protection coefficient. This accurately quantifies the essential differences in texture features between the two types of candidate regions, and also uses the global standard deviation to offset the interference of overall texture distribution fluctuations on the difference representation, ensuring that the coefficient truly reflects the uniqueness of the defect texture. This design avoids the problems of insufficient targeted defect texture protection and weak anti-interference ability caused by traditional fixed coefficients or single-dimensional feature calculations. It achieves adaptive adjustment of the protection coefficient, providing a reliable basis for subsequent accurate differentiation between missing coatings and reflective defects and protecting the true defect texture information, thereby improving the accuracy and robustness of defect detection and identification.

[0018] Preferably, the construction of the sparse penalty weights for each pixel includes: ;in, It is a pixel. Sparse penalty weights; It is a pixel. Gray-scale inhibition factor; It is a pixel. Texture feature indicators; It is the background consistency enhancement coefficient; It is the defect texture protection coefficient; It is a natural exponential function.

[0019] This method integrates background consistency enhancement coefficients and grayscale suppression factors, as well as defect texture protection coefficients and texture feature indices. It employs an exponential function to construct pixel-level sparse penalty weights. The first half strengthens the sparse penalty in background regions to improve background consistency, while the second half weakens the penalty in defect texture regions to protect true defect information, achieving adaptive and precise weight adjustment. This design avoids the problems of insufficient background suppression and excessive penalty for defect textures caused by traditional single-dimensional or fixed weights. It allows weights to dynamically adapt to image background characteristics and defect texture features, providing targeted weight support for subsequent sparse representation and defect extraction. This effectively improves the balance between background noise reduction and defect preservation in image processing, enhancing the accuracy and robustness of defect detection.

[0020] Preferably, the step of separating the missing paint region from the sparse anomaly matrix by combining texture feature indicators includes: calculating the average and standard deviation of the texture feature indicators of all pixels; setting dual adaptive thresholds: the missing paint determination threshold is the average plus 1.5 times the standard deviation, and the reflection determination threshold is the average minus 1.5 times the standard deviation, wherein the 1.5 times standard deviation can be adjusted according to the actual situation; matching the corresponding texture feature indicator value for the abnormal pixels marked as 1 in the sparse anomaly matrix; filtering the abnormal pixels: if the texture feature indicator reaches or exceeds the missing paint determination threshold, retain its 1 mark in the sparse anomaly matrix; if it is lower than or equal to the reflection determination threshold, set its 1 mark to 0 to eliminate reflection interference; after filtering, the sparse anomaly matrix only retains the 1 mark corresponding to the missing paint region, thereby separating the missing paint region from the sparse anomaly matrix.

[0021] Secondly, the present invention provides a composite plate ceramic coating quality inspection system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned composite plate ceramic coating quality inspection method is implemented.

[0022] By adopting the above technical solution, a computer program for the above-mentioned method for detecting the quality of ceramic coatings on composite panels is generated and stored in a memory so that it can be loaded and executed by a processor. A terminal device can then be made based on the memory and the processor for convenient use.

[0023] The beneficial effects of this invention are as follows: By defining a sliding window on the grayscale analysis image to extract local neighborhood information of pixels, and constructing a grayscale suppression factor based on the local grayscale contrast calculated from this local neighborhood information and the mean and standard deviation of the local grayscale contrast of all pixels, it is possible to accurately distinguish between normal coating areas and abnormal areas, including reflective areas and uncoated areas, laying the foundation for subsequent weight construction; by using the Sobel operator to calculate the gradient magnitude, and combining the average gradient magnitude within the sliding window with the normalized maximum and minimum values ​​to construct the texture feature index, it is possible to effectively calibrate the misjudgment that may occur by relying solely on local grayscale contrast, and enhance the distinguishability of the three types of areas; by calculating the basic regularization parameters of the robust principal component analysis algorithm, the background consistency enhancement coefficient, and the defect texture protection coefficient, and combining the grayscale suppression factor and texture feature index to construct a sparsity penalty... The weighting algorithm achieves a differentiated distribution with the highest weight for normal coating areas, moderate weight for reflective areas, and lowest weight for uncoated areas. A dynamic regularization parameter is obtained by weighting the globally fixed regularization parameter in the robust principal component analysis algorithm with this sparse penalty weight. The optimized robust principal component analysis objective function, constructed based on this dynamic regularization parameter, avoids the problem of traditional fixed regularization parameters failing to balance reflective suppression and uncoated area retention. The resulting low-rank background matrix and sparse anomaly matrix accurately separate normal coating areas from anomaly areas. Finally, combined with texture feature indicators, uncoated and reflective areas are separated from the sparse anomaly matrix, significantly reducing the false negative and false positive rates. This improves the accuracy and reliability of composite plate ceramic coating quality inspection, meeting the high-precision requirements for coating defect detection in industrial production, and ensuring the quality and safety of composite plate products. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a method for testing the quality of ceramic coatings on composite panels according to the present invention; Figure 2 To illustrate the feature comparison diagrams of different regions in a composite plate ceramic coating quality detection method of the present invention, a is the original image, b is the grayscale local contrast distribution diagram, and c is the texture feature index distribution diagram. Figure 3 To illustrate the comparison between the detection results of the present invention and the prior art in the quality detection method of ceramic coating of composite plate, a is the detection result of traditional RPCA and b is the detection result of the present invention. Detailed Implementation

[0025] 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, not all, of the embodiments of the present invention. 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.

[0026] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0027] This invention discloses a method for quality testing of ceramic coatings on composite panels, referring to... Figure 1 This includes steps S1 to S3: S1. Acquire surface images of the composite board using an industrial camera and perform preprocessing to obtain grayscale analysis images.

[0028] It should be noted that during the acquisition of surface images of composite panels by industrial cameras, on the one hand, electromagnetic interference from the industrial environment and noise from the camera sensor itself are inevitably introduced. These noises are superimposed on the surface images, blurring the edge details of each area and further increasing the difficulty of subsequent robust principal component analysis algorithms to separate low-rank backgrounds and sparse defects. On the other hand, due to factors such as uneven distribution of light intensity in industrial scenes, fluctuations in the flatness of the composite panel surface, and slight deviations in coating thickness, uneven illumination of the surface images is prone to occur. This leads to random fluctuations in the grayscale of normal ceramic coating areas, and the true grayscale characteristics of reflective and uncoated areas are masked. Therefore, surface images need to be preprocessed to obtain high-quality grayscale analysis images, laying a good foundation for subsequent accurate separation of low-rank backgrounds and sparse defects.

[0029] Specifically, the acquired color surface image is first divided into equal parts. Each region Take a positive integer, which can be set according to the image resolution. For example, Each region is individually subjected to adaptive histogram equalization. This process avoids noise amplification caused by global equalization and accurately eliminates global illumination shifts between different regions. It appropriately increases the grayscale values ​​of darker areas and decreases the grayscale values ​​of brighter areas within the normal coating region, ultimately achieving a uniform overall brightness distribution within the normal coating region. Furthermore, it eliminates only invalid illumination differences while fully preserving the grayscale differences between reflective areas, uncoated areas, and the normal coating region within the same area. Secondly, the surface image after adaptive histogram equalization is converted to grayscale to simplify the calculation process and improve processing efficiency. Next, Gaussian filtering is used to smooth the surface image. Specifically, a Gaussian smoothing template is constructed, and the gray value of each pixel in the surface image is calculated by weighted averaging. This can suppress high-frequency noise and eliminate the impact of random noise on image quality while preserving key details such as edge contours in the surface image, thus reducing the risk of false detection. Finally, the contrast enhancement processing is performed on the denoised surface image. A linear stretching method is used to adjust the gray dynamic range of the surface image, making the difference in gray values ​​between reflective areas, uncoated areas and normal coated areas more obvious, thereby obtaining a gray-scale analysis image.

[0030] It should be noted that adaptive histogram equalization avoids the noise amplification problem caused by global equalization, while accurately eliminating global illumination shifts between different regions: it appropriately increases the grayscale values ​​of darker areas in the normal coating area and appropriately decreases the grayscale values ​​of brighter areas in the normal coating area, ultimately resulting in a uniform overall brightness distribution in the normal coating area. It is important to note that this process only eliminates invalid illumination differences, completely preserving the grayscale differences between reflective and uncoated areas within the same region. Grayscale processing simplifies the calculation process and improves processing efficiency. Using Gaussian filtering to smooth the image can eliminate the impact of random noise on image quality while preserving surface image details, reducing the risk of false detections.

[0031] S2. Based on the local grayscale contrast of pixels in the grayscale analysis image, obtain the grayscale suppression factor of the pixels; based on the local gradient magnitude of the pixels, construct the texture feature index of the pixels; based on the grayscale suppression factor and the texture feature index of the pixels, construct the sparse penalty weight of the pixels as the final input to the RPCA algorithm.

[0032] It should be noted that the composite plate consists of a metal substrate and a ceramic coating. The ceramic coating mainly diffuses light, but there is some reflection. The grayscale distribution of the normal coating area in the composite plate is stable, and the local gradient changes are mild. Although the pre-processed grayscale analysis image has suppressed some noise and reflection interference, the image features of the reflective area and the uncoated area partially overlap, which will still directly affect the accuracy of subsequent ceramic coating quality detection. Specifically: the grayscale values ​​of the pixels in the reflective area are uniformly distributed and have high grayscale values, with low differences from the grayscale values ​​of the surrounding pixels, and the grayscale value changes are gentle with small local gradient amplitudes. The pixels in the uncoated area have obvious grayscale value differences from the pixels in the normal coating area, but because the area of ​​the uncoated area is small, the grayscale value difference between the uncoated area and the surrounding pixels is also low. Therefore, the image features such as grayscale distribution and local gradient of the two types of areas will partially overlap.

[0033] Specifically, to calculate the local grayscale contrast and local gradient magnitude of a pixel, it is necessary to first extract the local neighborhood information of each pixel using a sliding window. The specific operations include: defining a local neighborhood information on the grayscale analysis image. Sliding window of different sizes ,in, The value range is an odd number greater than 1 to ensure that the center of the sliding window is unique. The specific value can be set according to actual needs. Choose an odd number between 5 and 9, prioritizing the fit based on the fineness of the ceramic coating texture; the finer the texture, such as an ultra-thin coating, the better. Choose a smaller value to avoid over-smoothing and missing details; for coarser textures, such as thicker coatings... Taking a larger value suppresses texture noise interference, for example... Using the top-left pixel of the grayscale analysis image as the origin, the horizontal direction to the right is... The axis, vertically downward direction is A Cartesian coordinate system is constructed using axes to obtain the positions of all pixels in the grayscale analysis image and their corresponding grayscale values; using pixels as the basis for the analysis. Let's take a case study: [Analysis of pixel points] The grayscale value is denoted as , to pixel As a sliding window At the center, the sliding window... Contains pixels The total number of all pixels, including the total number of pixels, is denoted as . .

[0034] Furthermore, iterate through the sliding windows sequentially. Inner Pixels For each surrounding pixel other than the one in the surrounding pixels, the number of surrounding pixels is determined by the number of surrounding pixels. Taking this as an example for analysis, among which, and First based on pixels With surrounding pixels Given the positions of the two, calculate the Euclidean distance between them, denoted as . Reflecting pixels through Euclidean distance With surrounding pixels Spatial distance between them; then statistically analyzed by pixels. Centered sliding window Inside, pixels The standard deviation of the Euclidean distance to all surrounding pixels is denoted as . The Euclidean distance is normalized using this standard deviation to eliminate the influence of the sliding window size on the distance scale; based on this, the pixel point is calculated. The local grayscale contrast can be found in the following formula: ; in, It is a pixel. Local grayscale contrast; It is a pixel. grayscale value; It is the surrounding pixels grayscale value; It is the index of the pixel located at the center of the sliding window; It is the index of the peripheral pixels located outside the center of the sliding window; yes A sliding window of varying sizes, where... The value of is an odd number greater than 1; It is a pixel. With surrounding pixels The Euclidean distance between them; Based on pixels Centered sliding window Inside, pixels The standard deviation of the Euclidean distance from all surrounding pixels; It is a sliding window Contains pixels The total number of all pixels, including those in the array; It is a natural exponential function; It is a pixel. With surrounding pixels The difference in grayscale values; It is an absolute value function.

[0035] When pixel With surrounding pixels Gray value difference Larger, more pixels With surrounding pixels Euclidean distance between The smaller, in pixels Centered sliding window Inside, pixels Standard deviation of Euclidean distance from all surrounding pixels Larger, and sliding window Contains pixels The total number of all pixels, including The smaller the pixel, the more pixels The greater the local grayscale contrast, the lower the contrast, and vice versa.

[0036] It should be noted that there is an inherent difference in material grayscale between the uncoated area and the normally coated area. This feature can be directly reflected by the difference in grayscale value between the pixel at the center of the sliding window and its surrounding pixels; an exponential function is introduced into the relational formula. The algorithm assigns greater weight to pixels closer to the center of the sliding window. The physical meaning of this is that the properties of a pixel are more affected by its immediate surroundings. Normalizing the Euclidean distance using the standard deviation can eliminate the scale effect caused by different sliding window sizes, making the weight allocation more adaptive. The relation can accurately reflect the index of the degree of drastic change in local grayscale, namely the local grayscale contrast of the pixel.

[0037] Specifically, within the reflective area, the overall grayscale values ​​of all pixels are relatively high and the distribution is relatively uniform; therefore, the area located in the sliding window... center pixel grayscale values ​​and surrounding pixels Difference between gray values Overall, the pixel size is relatively small; although greater weight is given to pixels closer to the center of the sliding window, the local grayscale contrast is low after summing and averaging because the difference in grayscale values ​​is extremely small; in the unpainted area, the pixel size located within the sliding window... center pixel It is more likely to be located on a metal substrate with a higher grayscale value, while its surrounding pixels... Most of these areas will be located within the normal coating area where the grayscale value is lower. Therefore, although the area of ​​the missed coating may be small, there is a significant difference in grayscale value between it and the pixels in the adjacent normal coating area. The overall size is relatively large; and it assigns greater weight to the surrounding pixels that are closer to the pixel located at the center of the sliding window, ensuring that this significant difference in gray value can be highlighted in the calculation, resulting in a higher local gray value contrast after summing and averaging; in the normal coating area, the gray values ​​of all pixels are generally low and relatively uniform, and their local gray value contrast is usually between that of the reflective area and the uncoated area.

[0038] Furthermore, after obtaining the local grayscale contrast of all pixels, the mean and standard deviation of the local grayscale contrast of all pixels are calculated and denoted as follows: Pixel points are constructed based on the Gaussian kernel function. The grayscale suppression factor can be found in the following formula: ; in, It is a pixel. The grayscale suppression factor has a value range of . ; It is a pixel. Local grayscale contrast; It is the average local grayscale contrast of all pixels; It is the standard deviation of the local grayscale contrast of all pixels; It is a natural exponential function; when the pixel... The closer the local grayscale contrast is to the mean local grayscale contrast of all pixels, the better. When the difference is smaller, the standard deviation of the local grayscale contrast of all pixels is larger, and the pixel... The larger the grayscale inhibition factor, the closer it is to 1, and vice versa.

[0039] It should be noted that the detection scenario of this invention is based on actual industrial production images of ceramic coatings on composite panels. In such images, the grayscale distribution of normal coating areas, reflective areas, and areas with missing coatings inherently differs. Furthermore, industrial cameras retain minute texture details of the coating surface during acquisition, preventing the extreme case where all pixels have completely uniform local grayscale contrast. Therefore… It must be greater than 0; the formula does not have a denominator of 0. However, if this occurs due to a special test scenario... If the image is found to be defective, the ceramic coating of the composite plate corresponding to the image is determined to be free of defects and the result is treated as acceptable.

[0040] The method of mapping local grayscale contrast to a grayscale suppression factor using a Gaussian function has the following physical meaning: When the local grayscale contrast of a pixel is close to the average local grayscale contrast of all pixels, the larger its grayscale suppression factor (closer to 1) is, the more prevalent the feature of that pixel is. In this case, the pixel is determined to be in a normal coating area and needs to be suppressed in the subsequent RPCA algorithm. Conversely, when the local grayscale contrast of a pixel deviates from the average local grayscale contrast of all pixels, whether too high or too low, its grayscale suppression factor will be smaller (closer to 0), indicating that the feature of that pixel is most likely to be abnormal. In this case, the pixel is considered to be in a reflective area or a missing coating area and needs to be preserved in the subsequent RPCA algorithm. This processing can achieve less suppression for pixels in reflective areas with low local grayscale contrast and pixels in missing coating areas with high local grayscale contrast, while pixels in normal coating areas achieve greater suppression. This lays the foundation for subsequent fine differentiation between reflection and missing coating by combining texture features.

[0041] Specifically, the Sobel operator is used to calculate the gradient of the grayscale analysis image to obtain the gradient magnitude of each pixel. The Sobel operator is a well-known technique and will not be elaborated upon here. For example, the pixel-based gradient calculation method is used... Centered sliding window Calculate the average gradient magnitude of all pixels within the window, and then use the minimum-maximum normalization method to map this average value to... Range, as a pixel The texture feature index, denoted as The normalization benchmark is the maximum and minimum gradient magnitude of all pixels in the grayscale analysis image; by traversing all pixels, the texture feature index of each pixel is obtained.

[0042] It should be noted that in the reflective area, the grayscale distribution of pixels is uniform and the grayscale change rate is extremely low, so the gradient amplitude is extremely small and its texture feature index is close to 0; in the uncoated area, there is a boundary area between the metal substrate and the normal coating, the grayscale change is drastic, the gradient amplitude is extremely large, and its texture feature index is close to 1; in the normal coating area, the grayscale change is gradual, the gradient amplitude is at an intermediate level, and its texture feature index is between 0 and 1.

[0043] Texture feature indicators can effectively correct misjudgments that may occur if grayscale local contrast is relied upon alone. Specifically, some local texture abrupt changes in normal coating areas may cause their local grayscale contrast to deviate from the average local grayscale contrast of all pixels, leading to these pixels being misjudged as abnormal areas. The texture feature indicators of these pixels fall within the middle range between the texture feature indicators of reflective areas and uncoated areas. This middle range is determined based on sample statistics or experimental calibration in actual application scenarios. By using the range attribute of this texture feature indicator, the aforementioned misjudgments can be corrected, thereby further enhancing the distinction between reflective areas, uncoated areas, and normal coating areas.

[0044] Furthermore, the mean and standard deviation of the grayscale suppression factor for all pixels are calculated and denoted as follows: The normal threshold is set according to the material characteristics of the ceramic coating of the composite panel and the required testing accuracy. The normal threshold The range of values ​​is set to to Between, preferably, a normal threshold is set. Pixels with a grayscale suppression factor greater than or equal to the normal threshold are classified into a normal candidate set belonging to the normal coating region. The average grayscale suppression factor of all pixels in the normal candidate set is calculated and denoted as . Pixels with a grayscale suppression factor less than the normal threshold are classified into an abnormal candidate set, which includes reflective areas and areas with missing coating. The average grayscale suppression factor of all pixels in the abnormal candidate set is calculated and denoted as . Then, the background consistency enhancement coefficient is calculated, as shown in the following formula: ; in, It is the background consistency enhancement coefficient, which is used to amplify the weight difference between normal and abnormal regions, ensuring that normal regions are strongly suppressed. It is the average value of all grayscale suppression factors in the normal candidate set; It is the average value of all grayscale suppression factors in the abnormal candidate set; It is the standard deviation of the grayscale suppression factor for all pixels.

[0045] It should be noted that the detection object of this invention is a composite plate ceramic coating image containing both normal and abnormal areas. The distribution of the grayscale suppression factor is determined by the inherent differences between the normal and abnormal areas, and therefore inevitably exhibits dispersion. It is always greater than 0, thus avoiding the case where the denominator is 0; if it occurs under special working conditions... If the image shows no abnormal areas, it can be directly determined that the ceramic coating of the composite plate is of acceptable quality.

[0046] Furthermore, the standard deviation of the texture feature index of all pixels is obtained, denoted as . Next, for the anomaly candidate set belonging to the abnormal region, the texture feature index of each pixel in the set is obtained, and the anomaly threshold is calculated using the Otsu thresholding algorithm. Texture feature indicators greater than or equal to the anomaly threshold Pixels are divided into a candidate set of missing areas; texture feature indicators less than an abnormal threshold are considered. The pixels are divided into reflective candidate sets belonging to the reflective region; the average texture feature index of all pixels in the missing paint candidate set and the reflective candidate set is calculated and denoted as . Then, the defect texture protection coefficient is calculated, as shown in the following formula: ; in, It is a defect texture protection coefficient, used to amplify the weight difference between missed coating and reflection, ensuring that the weight of the missed coating area decays exponentially, that is, it obtains weaker suppression, while the reflective area obtains stronger suppression, thus achieving accurate differentiation between the two types of anomalies. It is the average of the texture feature indices of all pixels in the candidate set of missing paint. It is the average of the texture feature indices of all pixels in the reflective candidate set; It is the standard deviation of the texture feature index of all pixels.

[0047] It should be noted that in industrial images of composite ceramic coatings, the uncoated areas exhibit a significant gradient change at the interface between the metal substrate and the coating, while the reflective areas show no obvious gradient due to uniform grayscale. The texture feature indices of these two types of areas are fundamentally different, and the texture feature indices of the normally coated areas fall between the two. This inevitably leads to a discrete distribution of texture feature indices across all pixels. Always greater than 0, the formula has no risk of the denominator being 0; however, in special scenarios... If the image shows no areas with missing coating or reflective areas, then the quality of the ceramic coating on the composite panel meets the requirements.

[0048] Based on background consistency enhancement coefficient and defect texture protection coefficient And the grayscale suppression factor and texture feature index of each pixel are used to calculate the sparse penalty weight of each pixel. For example, the following relation can be used for analysis: ; in, It is a pixel. The larger the value of the sparsity penalty weight, the stronger the sparsity suppression strength of that pixel in the RPCA algorithm. It is a pixel. The larger the grayscale suppression factor, the more closely the grayscale value of the pixel matches the global distribution, and the more likely it is to be a normal area. It is a pixel. The larger the value of the texture feature index, the greater the gradient amplitude of the pixel, and the more likely it is to be a missed area; conversely, the smaller the gradient amplitude of the pixel, the more likely it is to be a reflective area. It is the background consistency enhancement coefficient; It is the defect texture protection coefficient; It is a natural exponential function.

[0049] For areas with normal coating, the core feature is the pixel. Gray-scale suppression factor Significantly large and pixel size Texture feature index It falls within the middle range of the overall statistical data, i.e. , Texture feature indices for all pixels The mean and standard deviation, at this time Item due to grayscale inhibition factor Significantly large values ​​will exhibit a strong amplification effect, while Item texture feature index Located in the middle range of the overall image statistics, therefore there is no significant attenuation, ultimately resulting in the pixel... Sparse penalty weights It is near the upper limit of the range, corresponding to the strongest sparsity suppression in the RPCA algorithm, ensuring that normal regions are judged as background.

[0050] For areas with missing coating, the core feature is the pixel. Texture feature index Significantly large, and pixel size Gray-scale suppression factor Significantly small, at this time Item due to texture feature index Significantly large, exhibiting exponential decay, while Item due to pixel Gray-scale suppression factor The pixel size is significantly too small, resulting in no noticeable magnification and ultimately causing pixel density issues. Sparse penalty weights It is near the lower limit of the value, which corresponds to the weakest sparsity suppression in the RPCA algorithm, ensuring that the missed areas are preserved.

[0051] For reflective areas, the core feature is the pixel. The smaller the texture feature index, i.e. the lower the gradient magnitude, and the smaller the gray-level suppression factor, the better. The item showed no significant decrease, and The item has no significant amplification effect, and the final sparse penalty weight is between the normal coating area and the uncoated area, balancing the need to suppress reflective interference and avoid suppressing weak defects.

[0052] Furthermore, iterate through all pixels to obtain the sparse penalty weights for all pixels.

[0053] S3. Weight the globally fixed regularization parameters by sparse penalty weights to obtain dynamic regularization parameters; construct an optimized objective function based on the dynamic regularization parameters, solve the objective function to obtain a sparse anomaly matrix; separate the missing coating areas from the sparse anomaly matrix by combining texture feature indicators, and output them as the quality inspection results of the ceramic coating of the composite plate.

[0054] It should be noted that step S2 calculates the sparse penalty weight for each pixel using local grayscale contrast and texture feature indices. Pixels in the normal coating area have the highest sparse penalty weight, those in the reflective area have a medium sparse penalty weight, and those in the uncoated area have the lowest coefficient penalty weight. Based on the sparse penalty weight of each pixel, the globally fixed regularization parameter in the Robust Principal Component Analysis (RPCA) algorithm is adjusted. The goal is to accurately separate the uncoated areas from the reflective areas, and ultimately output the quality inspection results for the uncoated areas that only contain the ceramic coating of the composite board.

[0055] Specifically, the grayscale analysis image of the ceramic coating of the composite plate obtained in step S1 is converted into a matrix form, denoted as the observation matrix. Step S2 yields the sparse penalty weights of all pixels and the observation matrix. Sparse penalty weight matrix of the same dimension Among them, the sparse penalty weight of pixels in the normal coating area is the largest, the sparse penalty weight of pixels in the reflective area is medium, and the sparse penalty weight of pixels in the uncoated area is the smallest.

[0056] Furthermore, the globally fixed regularization parameters With sparse penalty weight matrix The elements at corresponding positions are weighted to obtain the dynamic regularization parameters for each pixel. For example, the pixels are... The dynamic regularization parameter is denoted as Based on the dynamic regularization parameters of each pixel, an optimized objective function for robust principal component analysis is constructed.

[0057] Specifically, the objective function of the optimized robust principal component analysis is solved by combining existing robust principal component analysis techniques, and the observation matrix is... The matrix is ​​decomposed into a low-rank background matrix and a sparse anomaly matrix. The low-rank background matrix corresponds to the normal coating area of ​​the ceramic coating of the composite plate, i.e., the gray-level distribution pattern. The matrix exhibits low-rank characteristics. The sparse anomaly matrix is ​​a binary matrix, in which anomaly pixels are marked as 1 and background pixels are marked as 0. The anomaly pixels marked as 1 correspond only to the uncoated areas and reflective areas of the ceramic coating of the composite plate. The construction process is existing technology and will not be described in detail here.

[0058] It's important to note that the core reason for not directly using texture feature indicators to screen for missed areas in the ceramic coating of composite panels is the significant limitations of single texture feature indicators: they easily misclassify normal coatings with rough textures as missed areas and minor stains on smooth coatings as reflective areas, leading to a high false positive rate; furthermore, they can only describe pixel gradient amplitudes and cannot distinguish the essential difference between missed areas (true structural anomalies) and reflections (temporary imaging anomalies), resulting in insufficient robustness under different operating conditions. By first processing the grayscale analysis image using a robust principal component analysis algorithm, it can be decomposed into a low-rank background matrix corresponding to the normal coating and a sparse anomaly matrix containing only missed and reflective areas. This accurately delineates the range of abnormal pixels and eliminates interference from normal textures. Combining this with texture feature indicator screening avoids the shortcomings of single-feature screening, thereby improving detection accuracy.

[0059] Furthermore, combining the texture feature indices obtained in step S2, the following specific defect separation operations are performed: First, calculate the statistical characteristics of the texture feature indices for all pixels, namely the mean and standard deviation, and set dual adaptive thresholds: the missing coating threshold is the mean of the texture feature indices plus 1.5 times the standard deviation; features meeting this threshold condition are considered missing coating areas. The reflection threshold is the mean of the texture feature indices minus 1.5 times the standard deviation; features meeting this threshold condition are considered reflection areas. The 1.5 times standard deviation can be adjusted according to actual conditions; 1.5 is an example value and can be adjusted according to detection accuracy requirements. Second, establish a one-to-one correspondence between the sparse anomaly matrix and the pixel positions of the texture feature indices of the entire image. In the sparse anomaly matrix, each anomalous pixel marked as 1 is precisely matched with its corresponding texture feature index value. The third step involves filtering and retaining anomalous pixels: if the texture feature index value of an anomalous pixel reaches or exceeds the missing coating threshold and matches the characteristics of a missing coating area, then the pixel marked as 1 in the sparse anomaly matrix is ​​retained; if the texture feature index value of an anomalous pixel is lower than or equal to the reflection threshold and matches the characteristics of a reflection area, then the pixel marked as 1 in the sparse anomaly matrix is ​​set to 0, thus eliminating reflection interference. After the above filtering, only the 1 marker corresponding to the missing coating area is retained in the sparse anomaly matrix. This matrix is ​​then converted into a visual image, and the final output contains only the quality inspection results for the missing coating area of ​​the composite plate ceramic coating.

[0060] This invention also discloses a composite plate ceramic coating quality inspection system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement a composite plate ceramic coating quality inspection method according to the present invention.

[0061] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for quality inspection of ceramic coatings on composite panels, characterized in that, include: Obtain grayscale analysis images of the ceramic coating; Define a sliding window on the grayscale analysis image to extract the local neighborhood information of each pixel; Based on the local neighborhood information and the gray value of the pixel, the local gray contrast of each pixel is obtained; according to the distribution of the local gray contrast of all pixels, the gray suppression factor of each pixel is constructed; the normalized result of the average gradient magnitude of all pixels in the sliding window is used as the texture feature index of each pixel. The background consistency enhancement coefficient is obtained based on the difference in grayscale suppression factors between the normal candidate set and the abnormal candidate set, and the standard deviation of the grayscale suppression factors of all pixels. Based on the difference in texture feature indices between the candidate sets of missing coating and the candidate sets of reflection, and the standard deviation of texture feature indices of all pixels, a defect texture protection coefficient is obtained. Combined with the background consistency enhancement coefficient, grayscale suppression factor and texture feature indices, a sparse penalty weight for each pixel is constructed. The globally fixed regularization parameter is weighted by a sparse penalty weight to obtain a dynamic regularization parameter; an optimized objective function is constructed based on the dynamic regularization parameter, and the objective function is solved to obtain a sparse anomaly matrix; the uncoated areas are separated from the sparse anomaly matrix by combining texture feature indicators, and the results are used as the quality inspection results of the ceramic coating of the composite plate and output.

2. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The method for obtaining the local grayscale contrast of each pixel includes: ； in, It is a pixel. Local grayscale contrast; It is a pixel. grayscale value; It is the surrounding pixels grayscale value; It is the index of the pixel located at the center of the sliding window; It is the index of the peripheral pixels located outside the center of the sliding window; yes A sliding window of varying sizes, in which It is an odd number greater than 1; It is a pixel. With surrounding pixels The Euclidean distance between them; Based on pixels Centered sliding window Inside, pixels The standard deviation of the Euclidean distance from all surrounding pixels; It is a sliding window Contains pixels The total number of all pixels, including those in the array; It is a natural exponential function; It is an absolute value function.

3. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The step of constructing a grayscale suppression factor for each pixel based on the distribution of local grayscale contrast of all pixels includes: ； in, It is a pixel. Gray-scale inhibition factor; It is a pixel. Local grayscale contrast; It is the average local grayscale contrast of all pixels; It is the standard deviation of the local grayscale contrast of all pixels; It is a natural exponential function.

4. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The method for obtaining the normal candidate set and the abnormal candidate set includes: Calculate the mean and standard deviation of the grayscale suppression factor for all pixels, denoted as . The normal threshold is set according to the material characteristics of the ceramic coating of the composite panel and the required testing accuracy. The normal threshold The range of values ​​is set to to Pixels with a grayscale suppression factor greater than or equal to the normal threshold are classified as normal candidate sets belonging to the normal coating region; pixels with a grayscale suppression factor less than the normal threshold are classified as abnormal candidate sets belonging to the abnormal region.

5. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The step of obtaining the background consistency enhancement coefficient based on the difference in grayscale suppression factors between the normal candidate set and the abnormal candidate set, and the standard deviation of the grayscale suppression factors of all pixels, includes: ； in, It is the background consistency enhancement coefficient; It is the average value of all grayscale suppression factors in the normal candidate set; It is the average value of all grayscale suppression factors in the abnormal candidate set; It is the standard deviation of the grayscale suppression factor for all pixels.

6. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The method for obtaining the candidate set of missing coating and the candidate set of reflection includes: The standard deviation of the texture feature index of all pixels is obtained and denoted as . Next, for the anomaly candidate set belonging to the abnormal region, the texture feature index of each pixel in the anomaly candidate set is obtained, and the distribution of the texture feature index is used to calculate the anomaly threshold. Texture feature indicators greater than or equal to the anomaly threshold Pixels are divided into a candidate set of missing areas; texture feature indicators less than an abnormal threshold are considered. The pixels are divided into a set of reflective candidate points belonging to the reflective area.

7. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The step of obtaining the defect texture protection coefficient based on the difference in texture feature indices between the missing coating candidate set and the reflection candidate set, and the standard deviation of the texture feature indices of all pixels, includes: ； in, It is the defect texture protection coefficient; It is the average of the texture feature indices of all pixels in the candidate set of missing paint. It is the average of the texture feature indices of all pixels in the reflective candidate set; It is the standard deviation of the texture feature index of all pixels.

8. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The construction of the sparse penalty weights for each pixel includes: ； in, It is a pixel. Sparse penalty weights; It is a pixel. Gray-scale inhibition factor; It is a pixel. Texture feature indicators; It is the background consistency enhancement coefficient; It is the defect texture protection coefficient; It is a natural exponential function.

9. The method for quality inspection of ceramic coating on composite panels according to claim 1, characterized in that, The method of separating unpainted regions from the sparse anomaly matrix by combining texture feature indicators includes: Calculate the average and standard deviation of the texture feature indices for all pixels, and set dual adaptive thresholds: the missing paint determination threshold is the average plus 1.5 times the standard deviation, and the reflection determination threshold is the average minus 1.5 times the standard deviation. The 1.5 times standard deviation can be adjusted according to the actual situation. For abnormal pixels marked as 1 in the sparse anomaly matrix, match their corresponding texture feature index values. Filter abnormal pixels: if the texture feature index reaches or exceeds the missing paint determination threshold, retain its 1 mark in the sparse anomaly matrix; if it is lower than or equal to the reflection determination threshold, set its 1 mark to 0 to eliminate reflection interference. After filtering, the sparse anomaly matrix only retains the 1 mark corresponding to the missing paint area, thus separating the missing paint area from the sparse anomaly matrix.

10. A quality inspection system for ceramic coatings on composite panels, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a method for quality inspection of ceramic coatings on composite panels according to any one of claims 1-9.

Images