Appearance quality detection method and system for ALC plate

Abstract

The invention discloses an ALC plate appearance quality detection method and system, and belongs to the technical field of plate detection, and the method specifically comprises the steps: measuring the distance between a camera and a plate through a laser radar, enabling the camera to adjust the focusing point of a lens according to the distance fed back by the laser radar, enabling a focusing plane to coincide with the surface of the plate, and adjusting the depth of field of the lens to be shallow depth of field; preprocessing the surface image of the plate, identifying a confusion circle in the preprocessed surface image, acquiring diameter data of the confusion circle, and checking whether the variance of the diameter data meets a preset requirement or not; for unqualified plates, acquiring node diameters of the diameter data, screening out the diameter data larger than the node diameters, marking corresponding dispersion circles as abnormal dispersion circles, and marking the minimum bounding rectangle of the abnormal dispersion circles as a target area; the depth of field is adjusted to be deep depth of field, the plate surface image is obtained again, an image area corresponding to the target area is cut off, the image area is input into a preset model, and a corresponding type result is output.

Description

Technical Field

[0001] This invention relates to the field of sheet metal testing technology, specifically to a method and system for testing the appearance quality of ALC sheets. Background Technology

[0002] In modern industrial production, automated inspection technology has become an important means to improve production efficiency and ensure product quality. This is especially true in the building materials industry, such as on ALC (Autoclaved Lightweight Concrete) panel production lines, where appearance quality inspection is crucial. As a lightweight, high-strength building material, the surface quality of ALC panels directly affects the product's aesthetics and performance. Traditional manual inspection methods are not only inefficient but also susceptible to human factors, such as operator fatigue and subjective judgment, leading to inconsistent results. This inconsistency may cause minor defects to be overlooked or normal products to be misjudged as substandard, thus impacting the efficiency of the entire production line and the overall product quality. Therefore, developing an efficient and accurate automated inspection method for the appearance quality of ALC panels is of significant practical importance and can provide strong technical support for quality control in the building materials industry.

[0003] Currently, several automated inspection systems based on image processing technology exist on the market. These systems typically rely on full-image capture by high-resolution cameras for feature extraction and analysis. High-resolution cameras can capture subtle features on the surface of the material, providing rich information for subsequent image processing. However, the number and complexity of features contained in a full-image photograph increase computational complexity and time costs. Due to the massive amount of image data to be processed and the high complexity of the algorithms, the response time of the inspection system is long, failing to meet the demands of rapid production. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for inspecting the appearance quality of ALC (Alternating Current) sheets, and to solve the following technical problems: The number of features contained in the entire photo is huge and complex, and performing image recognition on the entire photo will increase the computational complexity and time cost.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for inspecting the appearance quality of ALC boards includes the following steps: The sheet material is moved under the industrial camera, and the distance between the camera and the sheet material is measured by LiDAR. Based on the distance fed back by the LiDAR, the camera adjusts the focus point of the lens so that the focusing plane coincides with the surface of the sheet material, and adjusts the depth of field of the lens to a shallow depth of field. The surface image of the board is obtained by cutting and preprocessing. The diffuse circle in the preprocessed surface image is identified and the diameter data of the diffuse circle is obtained. The variance of the diameter data is checked to see if it meets the preset requirements. If it does, it means that the surface flatness of the board is qualified; if not, it means that the surface flatness of the board is unqualified. For substandard boards, statistical analysis is performed on the diameter data to obtain the node diameters of the diameter data. Diameter data larger than the node diameters are filtered out, and the corresponding dispersion circles are marked as abnormal dispersion circles. The smallest bounding rectangle of the abnormal dispersion circles is marked as the target area, and overlapping target areas are merged. Adjust the lens depth of field to deep depth of field, acquire the image of the board surface again, and crop the image area corresponding to the target area. Input the image area into the preset model and output the corresponding unevenness type result.

[0006] As a further aspect of the present invention: the process of preprocessing the surface image, identifying the blur circles in the preprocessed surface image, and obtaining the diameter of the blur circles is as follows: The surface image is converted to a grayscale image through grayscale processing. The grayscale image is then subjected to Gaussian filtering. The Canny edge detection algorithm is used to detect edges in the image. The Sobel operator is used to calculate the gradient magnitude and direction of the image. The gradient magnitude represents the edge strength at each pixel, and the gradient direction represents the edge direction. By comparing the gradient magnitude and direction, the gradient magnitude of non-edge points is suppressed, and local maximum points on the edges are retained. High and low thresholds are set. The high threshold is used to determine strong edges, and the low threshold is used to determine weak edges. Strong edges are directly retained. By tracking the connectivity of the edges, weak edges are connected to strong edges to obtain a complete edge image. The Hough circle transform is used to map circles in image space to parameter space. For each edge point, the center position and radius range are determined according to its gradient direction. In parameter space, each center position corresponds to an accumulator. For each edge point, voting is performed in the accumulator based on the center position and radius. The value of the accumulator represents the number of edge points of the circle corresponding to that center position and radius. An accumulator threshold is set. When the value of the accumulator is greater than the threshold, it means that the circle corresponding to that center position and radius is valid. Based on the value of the accumulator, the center position and diameter of the detected circle are determined.

[0007] As a further aspect of the present invention, the accumulator threshold is set according to camera parameters, object distance, and plate size.

[0008] The specific process of obtaining the node diameters by performing statistical analysis on the diameter data is as follows: Sort the diameter data from smallest to largest to generate a diameter sequence D1, D2, ..., D...n Let n represent the number of diameter data points. An empty set is set up, and the diameter data points are added to this set sequentially. After each addition, the variance of all diameter data points in the current set is calculated. For the i-th diameter data point D... i Let i∈(1,n), calculate D1, D2, ..., D i variance C i , and D1, D2, ..., D i+1 variance C i+1 Calculate the variance C i With variance C i+1 The difference between them, if the difference is greater than a preset threshold, then D will be... i The difference is marked as the node diameter. If the difference is less than a preset threshold, the detection continues.

[0009] As a further aspect of the present invention, the process of merging overlapping target regions is as follows: A rectangular coordinate system is established with the top left corner of the plate surface image as the origin. The coordinates of the top left corner endpoints (x1, y1) and (x2, y2) of the two overlapping target areas are obtained respectively, and the coordinates of the bottom right corner endpoints (z1, w1) and (z2, w2) of the two target areas are obtained respectively. The coordinates of the upper left corner of the merged target region are marked as (X, Y), where X = min(x1, x2) and Y = min(y1, y2); the coordinates of the upper left corner of the merged new target region are marked as (Z, W), where Z = max(z1, z2) and W = max(w1, w2). Merge all overlapping target regions sequentially until all target regions exist independently.

[0010] As a further aspect of the present invention: priority is assigned to the target region, where the priority is the diameter of the dispersion circle, and the priority of the merged template region is the average value of the dispersion circle diameter within the region.

[0011] As a further aspect of the present invention: the lidar and the industrial camera are based on synchronized time coding, and the time difference between the distance data collected by the lidar and the image data collected by the industrial camera is constant.

[0012] This invention also includes an ALC sheet appearance quality inspection system, applied to the above-mentioned ALC sheet appearance quality inspection method, comprising: LiDAR is used to measure the distance between the camera and the material, and send the distance data to the industrial camera. An industrial camera is used to adjust the focus point of the lens based on the distance fed back by the lidar, so that the focusing plane coincides with the surface of the material, and to adjust the depth of field of the lens to a shallow depth of field or a deep depth of field. The preprocessing module is used to cut and obtain a shallow depth-of-field image of the board surface, preprocess the surface image, identify the circle of confusion in the preprocessed surface image, obtain the diameter data of the circle of confusion, and check whether the variance of the diameter data meets the preset requirements. If yes, it means that the flatness of the board surface is qualified; if no, it means that the flatness of the board surface is unqualified. The target selection module is used to perform statistical analysis on the diameter data of unqualified boards, obtain the node diameter of the diameter data, filter out the diameter data larger than the node diameter, mark the corresponding dispersion circle as abnormal dispersion circle, mark the minimum bounding rectangle of the abnormal dispersion circle as the target area, and merge the target areas that overlap. The result detection module is used to acquire the surface image of the board material under deep depth of field, crop the image area corresponding to the target area, input the image area into a preset model, and output the corresponding unevenness type result.

[0013] The beneficial effects of this invention are: (1) This invention is based on the detection principle of the circle of confusion. It utilizes the characteristic that the farther the imaging point on the off-focus plane is from the focusing plane, the larger the corresponding circle of confusion. By screening larger circles of confusion, defects such as dents, cracks, or protrusions on the surface of the board are identified. By accurately measuring and screening larger circles of confusion, small defects such as dents, cracks, or protrusions on the surface of the board can be effectively identified, improving the accuracy of defect identification and ensuring the reliability of the detection results. After initially screening out larger circles of confusion, only image cropping and model recognition of the corresponding target area are required, reducing the amount of image data that needs to be processed, thereby reducing the system's computing power requirements, improving detection efficiency, and reducing hardware costs.

[0014] (2) This invention can quickly locate defect areas on the surface of the board, reducing the time required for point-by-point inspection of the entire board surface, improving overall inspection efficiency, and helping to achieve large-scale, high-efficiency board quality inspection. By marking abnormal dispersion circles and target areas, it provides clear guidance for subsequent defect handling. This invention does not depend on specific board types or defect morphologies, and can adapt to ALC boards of different specifications and types, as well as surface defects of various shapes and sizes, and has wide applicability and good adaptability. Attached Figure Description

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

[0016] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

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

[0018] Please see Figure 1 As shown, this invention provides a method for inspecting the appearance quality of ALC (Alternating Current Carbide) sheets, comprising the following steps: 1. In this ALC sheet appearance quality inspection method, after moving the sheet under the industrial camera, the distance between the camera and the sheet is first measured using a LiDAR. The LiDAR and industrial camera are based on synchronized time coding, meaning they acquire data synchronously, and the time difference between the distance data acquired by the LiDAR and the image data acquired by the industrial camera is constant. This synchronization mechanism ensures a precise correspondence between distance measurement and image acquisition, providing reliable basic data for subsequent image processing and defect detection.

[0019] Based on the distance information fed back by the LiDAR, the camera automatically adjusts the lens's focus point to align the focal plane with the surface of the material. This ensures the highest possible image clarity of the captured material surface, providing high-quality image data for subsequent image analysis and defect identification. Simultaneously, the camera adjusts the lens depth of field to a shallow depth of field. This shallow depth of field setting serves the following purposes: Highlighting dents or protrusions on the board surface: Shallow depth of field results in a very narrow focal plane for the camera, ensuring only the board surface remains in sharp focus while areas outside the surface gradually become blurred. Thus, when dents or protrusions exist on the board surface, these defective areas will exhibit a noticeable blurring effect due to their varying distance from the focal plane, creating a sharp contrast with the surrounding sharp board surface. This contrast makes it easier to identify and locate defects such as dents or protrusions on the board surface.

[0020] Enhancing the visual effect of defect features: Shallow depth of field makes the blurriness of the defect area proportional to its height variation; the greater the height variation, the more pronounced the blurriness. This visual effect enhances the recognizability of defect features, making defects stand out more in the image and facilitating subsequent image processing algorithms for detection and analysis.

[0021] Improving defect detection accuracy: Because shallow depth of field can clearly highlight minute height variations on the surface of the board, it can improve the detection accuracy of defects such as dents or protrusions on the board surface. Even tiny defects can be accurately identified through their blurred features in the image, thereby improving overall detection performance.

[0022] 2. Obtain a surface image of the board by cutting, preprocess the surface image, and identify the circles of confusion in the preprocessed surface image. The specific process is as follows: Grayscale processing: Converting a colored image of the board surface into a grayscale image. Grayscale images contain only brightness information, removing color information, simplifying image data, reducing computational complexity, and preserving the main structural features of the image, providing a foundation for subsequent operations such as edge detection and circle detection.

[0023] Gaussian filtering: Applying Gaussian filtering to grayscale images. Gaussian filtering is a smoothing filter that, through convolution with the image, can effectively remove noise, smooth image details, reduce false positives and false negatives in edge detection, and improve the robustness of image processing.

[0024] Edge detection and gradient calculation: Canny edge detection: The Canny edge detection algorithm is used to detect edges in Gaussian filtered images. The Canny algorithm is a multi-stage edge detection algorithm that accurately detects edges in an image through gradient calculation, non-maximum suppression, and hysteresis thresholding, while suppressing noise and false edges to obtain clear, continuous edge images.

[0025] The Sobel operator calculates gradients: The Sobel operator is used to calculate the magnitude and direction of the gradient in an image. A commonly used edge detection operator, the Sobel operator calculates the gradient in the horizontal and vertical directions of the image, obtaining the gradient magnitude and direction at each pixel. The gradient magnitude represents the edge strength at that pixel, i.e., the prominence of the edge; the gradient direction represents the direction of the edge, i.e., the direction of the edge. By comparing the gradient magnitude and direction, the gradient magnitude of non-edge points can be suppressed, while local maxima on edges can be preserved, thereby further optimizing the edge detection results.

[0026] High and low threshold settings: These settings differentiate between strong and weak edges in an image. A high threshold identifies strong edges (edges with large gradient magnitudes), while a low threshold identifies weak edges (edges with small gradient magnitudes). Strong edges are preserved directly because they are typically prominent features in the image; weak edges require further processing.

[0027] Edge connectivity tracking and connection: By tracking edge connectivity, weak edges are connected to strong edges. Specifically, if a weak edge point is connected to a strong edge point and their gradient directions are consistent, then the weak edge point is considered valid and connected to the strong edge. This yields a complete edge image, avoiding edge breaks and discontinuities, and improving the integrity of edge detection.

[0028] Hough Circle Transform: The Hough Circle Transform maps circles in image space to parameter space. It's a parameter space transformation method that detects circles in an image by mapping each edge point to a conic section in the parameter space. For each edge point, the center position and radius are determined based on its gradient direction. In parameter space, each center position corresponds to an accumulator. For each edge point, a vote is taken within the accumulator based on the center position and radius; the accumulator value represents the number of edge points corresponding to that center position and radius.

[0029] When using the Hough circle transform for circle detection, the accumulator threshold is a crucial parameter that determines which circles will be detected. The accumulator threshold setting needs to consider the number and size of circles in the image to ensure the accuracy and robustness of the detection results. Specifically: The effect of the number of circles on the accumulator threshold: A large number of circles: When there are many circles in an image, if the accumulator threshold is set too high, some smaller circles or circles with weak edges may not be detected because their accumulator values ​​may not have reached the threshold. In this case, the accumulator threshold needs to be appropriately reduced in order to detect more circles.

[0030] When there are few circles in an image, if the accumulator threshold is set too low, some non-circular edge structures may be falsely detected as circles because their accumulator values ​​may also have reached the threshold. In this case, the accumulator threshold needs to be appropriately increased to reduce false detections and ensure that the detected circles are true circles.

[0031] The effect of the circle size on the accumulator threshold: Larger radius circles: For circles with larger radii, there are more edge points, resulting in a relatively higher accumulation threshold. In this case, the accumulator threshold can be set relatively high because larger circles are more likely to obtain higher accumulated values ​​in the accumulator.

[0032] Smaller radius circles: For circles with smaller radii, the number of edge points is less, and the accumulation threshold is relatively low. In this case, the accumulator threshold needs to be set relatively low to detect the smaller circle. Otherwise, the smaller circle may be missed due to insufficient accumulation resistance.

[0033] The camera parameters and object distance determine the size of the circle of confusion radius, while the material size determines the number of circles of confusion. Therefore, the accumulator threshold is set based on the camera parameters, object distance, and material size to reduce false detections and false negatives.

[0034] Diameter Data Acquisition and Variance Check: Acquire the diameter data of the detected dispersion circles and check whether the variance of these diameter data meets the preset requirements. Variance is a statistic that measures the dispersion of data. If the variance of the diameter data is small, it indicates that the diameter distribution of the dispersion circles is relatively concentrated, and the surface flatness of the board is good; conversely, if the variance is large, it indicates that the diameter distribution of the dispersion circles is relatively dispersed, and the surface flatness of the board is poor.

[0035] Flatness assessment: Based on the variance check results, determine whether the surface flatness of the board is up to standard. If the variance of the diameter data meets the preset requirements, it indicates that the surface flatness of the board is up to standard; if the variance does not meet the requirements, it indicates that the surface flatness of the board is not up to standard, and there may be dents, bumps or other unevenness, requiring further processing or repair.

[0036] 3. For substandard boards, in-depth statistical analysis of the diameter data is required to identify abnormal dispersion circles and determine the target area: Data Sorting and Sequence Generation: First, sort all the obtained diameter data in ascending order to generate a diameter sequence: D1, D2, ..., D n The purpose of this step is to facilitate subsequent variance calculation and node diameter identification. By sorting, the distribution of diameter data can be visually observed, providing a basis for analyzing the dispersion of the data and identifying outliers.

[0037] Set initialization and variance calculation: A blank set is created to store the diameter data. The diameter data is then added to the set in sorted order. After each diameter data point is added, the variance of all diameter data in the current set is calculated. Variance is an important indicator of data dispersion; calculating variance helps understand the data distribution and determine whether the data is concentrated or dispersed.

[0038] Variance calculation and difference comparison: For the i-th diameter data D i Calculate from D1 to D i variance C i and from D1 to D i+1 variance C i+1 Variance is the square root of the variance, and it is also used to measure the dispersion of data. Calculate the variance C. i With variance C i+1 The difference between them, this difference reflects the addition of D i+1 The change in the dispersion of the data can be observed. By comparing the differences, the stability of the data distribution can be determined.

[0039] Node diameter identification: If variance C i With variance C i+1 If the difference between them is greater than a preset threshold, then D will be...i The node diameter is a key point in the diameter data distribution, marking the location where the data dispersion changes significantly. By identifying the node diameter, the diameter data can be divided into different intervals, providing a basis for subsequent screening of abnormal circles of dispersion; if the difference is less than a preset threshold, the detection continues until all diameter data has been processed.

[0040] Screening for anomalous dispersion circles: Based on the node diameter, filter out dispersion circles with diameters larger than the node diameter. These dispersion circles are considered anomalous because their diameters are significantly larger than other dispersion circles, potentially representing larger defects on the board surface, such as large dents or protrusions. By screening for anomalous dispersion circles, the focus of inspection can be placed on defects that may have a significant impact on the quality of the board.

[0041] Marking the target area of ​​the abnormal dispersion circle: The smallest bounding rectangle of the abnormal dispersion circle is marked as the target area. The smallest bounding rectangle is the smallest rectangle that can completely contain the dispersion circle. By marking the target area, the specific location on the board surface that needs further processing or analysis can be determined. Marking the target area provides a clear area range for subsequent defect processing or analysis, which helps to improve the targeting and efficiency of the processing.

[0042] Coordinate system establishment and coordinate acquisition: A rectangular coordinate system is established with the top left corner of the board surface image as the origin. Within this system, the coordinates of the top left corner endpoints (x1, y1) and (x2, y2) and the bottom right corner endpoints (z1, w1) and (z2, w2) of the two overlapping target regions are acquired. This coordinate acquisition provides accurate positional information for subsequent target region merging operations, ensuring the accuracy and rationality of the merging process.

[0043] Target Region Merging: For overlapping target regions, a merging operation is performed. The coordinates of the top-left endpoint of the merged target region are marked as (X, Y), where X = min(x1, x2) and Y = min(y1, y2); the coordinates of the bottom-right endpoint of the merged new target region are marked as (Z, W), where Z = max(z1, z2) and W = max(w1, w2). This method merges multiple overlapping target regions into a larger region, simplifying subsequent processing while ensuring that all abnormal diffusion circles are included. The merging operation helps reduce redundant processing steps and improves overall detection efficiency.

[0044] The merging operation is executed iteratively: all overlapping target areas are merged sequentially until all target areas exist independently and no longer overlap. This step ensures that the final target areas are non-overlapping, providing clear and accurate area division for subsequent defect processing or analysis. By repeatedly executing the merging operation, the problem of overlapping target areas can be comprehensively resolved, ensuring the accuracy and completeness of the detection results.

[0045] 4. Adjust the lens depth of field to deep depth of field: After identifying the abnormal circle of confusion and the target area, adjust the industrial camera lens depth of field to deep depth of field. Deep depth of field means that the camera's focusing range is expanded, so not only are defects within the target area clearly visible, but the surface of the board material within a certain range around the target area will also be in sharp focus. The purpose of this is to obtain more comprehensive image information of the board material surface, ensuring that other possible minor defects in and around the target area can be captured, providing richer data support for subsequent detailed analysis.

[0046] Acquire another image of the board surface: After adjusting to deep depth of field, photograph the board surface again to acquire a new image. This new image will contain clear image information of the target area and its surroundings, laying the foundation for subsequent target area cropping and analysis.

[0047] Image cropping of the target area: Based on the previously marked coordinates of the target area, the corresponding target area image is cropped from the acquired image of the board surface. The cropped image only contains the target area and a certain range around it, removing other irrelevant image information, making subsequent analysis more efficient and accurate.

[0048] Priority Assignment: Priority is assigned to the trimmed target areas. The priority is based on the diameter of the dispersion circle; a larger diameter indicates a more severe defect and a higher priority. For the merged template area, its priority is the average diameter of the dispersion circles within the area. The purpose of priority assignment is to rationally arrange the analysis order and resources according to the priority in subsequent model analysis, ensuring the rapid identification and handling of severe defects.

[0049] Input Model Analysis: Based on priority, the cropped target area images are sequentially input into a preset model. The preset model is a trained deep learning model or other machine learning model capable of identifying and classifying the types of unevenness on the board surface, such as dents, bumps, and cracks.

[0050] Output Unevenness Type Results: After analyzing the input target area image, the model outputs the corresponding unevenness type results. The results will detail the defect types within the target area, such as specific depression depths, protrusion heights, and crack lengths. These results provide accurate information for subsequent defect handling and quality improvement, helping to implement targeted repair measures and improve the overall quality of the board material.

[0051] This invention also includes an ALC sheet appearance quality inspection system, applied to the above-mentioned ALC sheet appearance quality inspection method, comprising: LiDAR is used to measure the distance between the camera and the material, and send the distance data to the industrial camera. An industrial camera is used to adjust the focus point of the lens based on the distance fed back by the lidar, so that the focusing plane coincides with the surface of the material, and to adjust the depth of field of the lens to a shallow depth of field or a deep depth of field. The preprocessing module is used to cut and obtain a shallow depth-of-field image of the board surface, preprocess the surface image, identify the circle of confusion in the preprocessed surface image, obtain the diameter data of the circle of confusion, and check whether the variance of the diameter data meets the preset requirements. If yes, it means that the flatness of the board surface is qualified; if no, it means that the flatness of the board surface is unqualified. The target selection module is used to perform statistical analysis on the diameter data of unqualified boards, obtain the node diameter of the diameter data, filter out the diameter data larger than the node diameter, mark the corresponding dispersion circle as abnormal dispersion circle, mark the minimum bounding rectangle of the abnormal dispersion circle as the target area, and merge the target areas that overlap. The result detection module is used to acquire the surface image of the board material under deep depth of field, crop the image region corresponding to the target area, input the image region into a preset model, and output the corresponding unevenness type result. The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for inspecting the appearance quality of ALC (Alternating Current Carbide) sheets, characterized in that, Includes the following steps: The sheet material is moved under the industrial camera, and the distance between the camera and the sheet material is measured by LiDAR. Based on the distance fed back by the LiDAR, the camera adjusts the focus point of the lens so that the focusing plane coincides with the surface of the sheet material, and adjusts the depth of field of the lens to a shallow depth of field. The surface image of the board is obtained by cutting and preprocessing. The diffuse circle in the preprocessed surface image is identified and the diameter data of the diffuse circle is obtained. The variance of the diameter data is checked to see if it meets the preset requirements. If it does, it means that the surface flatness of the board is qualified; if not, it means that the surface flatness of the board is unqualified. For substandard boards, statistical analysis is performed on the diameter data to obtain the node diameters of the diameter data. Diameter data larger than the node diameters are filtered out, and the corresponding dispersion circles are marked as abnormal dispersion circles. The smallest bounding rectangle of the abnormal dispersion circles is marked as the target area, and overlapping target areas are merged. Adjust the lens depth of field to deep depth of field, acquire the image of the board surface again, and crop the image area corresponding to the target area. Input the image area into the preset model and output the corresponding unevenness type result.

2. The method for inspecting the appearance quality of ALC sheet according to claim 1, characterized in that, The process of preprocessing the surface image, identifying the circle of confusion in the preprocessed surface image, and obtaining the diameter of the circle of confusion is as follows: The surface image is converted to a grayscale image through grayscale processing. The grayscale image is then subjected to Gaussian filtering. The Canny edge detection algorithm is used to detect edges in the image. The Sobel operator is used to calculate the gradient magnitude and direction of the image. The gradient magnitude represents the edge strength at each pixel, and the gradient direction represents the edge direction. By comparing the gradient magnitude and direction, the gradient magnitude of non-edge points is suppressed, and local maximum points on the edges are retained. High and low thresholds are set. The high threshold is used to determine strong edges, and the low threshold is used to determine weak edges. Strong edges are directly retained. By tracking the connectivity of the edges, weak edges are connected to strong edges to obtain a complete edge image. The Hough circle transform is used to map circles in image space to parameter space. For each edge point, the center position and radius range are determined according to its gradient direction. In parameter space, each center position corresponds to an accumulator. For each edge point, voting is performed in the accumulator based on the center position and radius. The value of the accumulator represents the number of edge points of the circle corresponding to that center position and radius. An accumulator threshold is set. When the value of the accumulator is greater than the threshold, it means that the circle corresponding to that center position and radius is valid. Based on the value of the accumulator, the center position and diameter of the detected circle are determined.

3. The method for inspecting the appearance quality of ALC sheet according to claim 1, characterized in that, The accumulator threshold is set based on camera parameters, object distance, and plate size.

4. The method for inspecting the appearance quality of ALC sheet according to claim 1, characterized in that, The specific process of obtaining the node diameters by performing statistical analysis on the diameter data is as follows: Sort the diameter data from smallest to largest to generate a diameter sequence D1, D2, ..., D... n Let n represent the number of diameter data points. An empty set is set up, and the diameter data points are added to this set sequentially. After each addition, the variance of all diameter data points in the current set is calculated. For the i-th diameter data point D... i Let i∈(1,n), calculate D1, D2, ..., D i variance C i , and D1, D2, ..., D i+1 variance C i+1 Calculate the variance C i With variance C i+1 The difference between them, if the difference is greater than a preset threshold, then D will be... i The difference is marked as the node diameter. If the difference is less than a preset threshold, the detection continues.

5. The method for inspecting the appearance quality of ALC sheet according to claim 1, characterized in that, The process of merging overlapping target regions is as follows: A rectangular coordinate system is established with the top left corner of the plate surface image as the origin. The coordinates of the top left corner endpoints (x1, y1) and (x2, y2) of the two overlapping target areas are obtained respectively, and the coordinates of the bottom right corner endpoints (z1, w1) and (z2, w2) of the two target areas are obtained respectively. The coordinates of the upper left corner of the merged target region are marked as (X, Y), where X = min(x1, x2) and Y = min(y1, y2); the coordinates of the upper left corner of the merged new target region are marked as (Z, W), where Z = max(z1, z2) and W = max(w1, w2). Merge all overlapping target regions sequentially until all target regions exist independently.

6. The method for inspecting the appearance quality of ALC sheet according to claim 1, characterized in that, The target region is assigned a priority, which is the diameter of the dispersion circle. The priority of the merged template region is the average diameter of the dispersion circle within the region.

7. The method for inspecting the appearance quality of ALC sheet according to claim 1, characterized in that, The lidar and industrial camera are based on synchronized time coding, and the time difference between the distance data collected by the lidar and the image data collected by the industrial camera is constant.

8. An appearance quality inspection system for ALC panels, applied to the appearance quality inspection method for ALC panels according to any one of claims 1-7, characterized in that, include: LiDAR is used to measure the distance between the camera and the material, and send the distance data to the industrial camera. An industrial camera is used to adjust the focus point of the lens based on the distance fed back by the lidar, so that the focusing plane coincides with the surface of the material, and to adjust the depth of field of the lens to a shallow depth of field or a deep depth of field. The preprocessing module is used to cut and obtain a shallow depth-of-field image of the board surface, preprocess the surface image, identify the circle of confusion in the preprocessed surface image, obtain the diameter data of the circle of confusion, and check whether the variance of the diameter data meets the preset requirements. If yes, it means that the flatness of the board surface is qualified; if no, it means that the flatness of the board surface is unqualified. The target selection module is used to perform statistical analysis on the diameter data of unqualified boards, obtain the node diameter of the diameter data, filter out the diameter data larger than the node diameter, mark the corresponding dispersion circle as abnormal dispersion circle, mark the minimum bounding rectangle of the abnormal dispersion circle as the target area, and merge the target areas that overlap. The result detection module is used to acquire the surface image of the board material under deep depth of field, crop the image area corresponding to the target area, input the image area into a preset model, and output the corresponding unevenness type result.

Images