Machine vision-based method for rapid detection of production quality of building decoration plates

Abstract

The application relates to the technical field of building decoration plate production detection, and discloses a building decoration plate production quality rapid detection method based on machine vision. The method first collects surface images, edge images and texture feature sequences of each visual detection unit in each detection cycle; then, based on the related deviation of abnormal points in the texture feature sequence, the feature deviation index of the abnormal points is obtained, and then the average deviation index and the quality fluctuation degree of the quality fluctuation collection moment are obtained to determine the surface quality coefficient; meanwhile, the edge quality coefficient is obtained based on the surface and edge images; finally, the detection parameter adjustment value is obtained according to the surface and edge quality coefficients, which is used for correcting the detection parameters and threshold values of the next cycle template matching defect identification. The method realizes dynamic adjustment of the detection parameters and improves the accuracy of defect identification.

Description

Technical Field

[0001] This invention relates to the field of building decorative panel production and testing technology, specifically a rapid testing method for the production quality of building decorative panels based on machine vision. Background Technology

[0002] In the production process of architectural decorative panels, product quality inspection is a crucial step in ensuring their market competitiveness. As the construction industry continues to raise its requirements for the appearance and performance of decorative materials, traditional manual inspection methods are gradually becoming insufficient to meet the demands of large-scale industrial production. Manual inspection not only relies on the experience and responsibility of the inspectors, but also suffers from fatigue during long hours of work, leading to decreased inspection efficiency and increased error rates. This is especially true when decorative panels have complex surface textures and diverse defect types, significantly reducing the accuracy of manual identification.

[0003] Currently, machine vision-based inspection technology has been applied in multiple industrial sectors and is gradually being promoted in the inspection of architectural decorative panels. However, existing machine vision inspection methods still have many limitations in practical applications. Most inspection systems use fixed detection parameters and thresholds for defect identification, making it difficult to adapt to quality fluctuations caused by changes in material properties and environmental factors during the production process. For example, in the production of decorative panels, factors such as batch differences in raw materials, fluctuations in production line speed, and changes in lighting conditions can all cause the same type of defect to present different characteristics in the image, making template matching under fixed parameters prone to missed detections or false detections.

[0004] Current technologies for analyzing the surface texture of decorative panels often focus on extracting and judging single features, neglecting the correlation between texture features and the differences between local and overall textures. Texture defects on the surface of decorative panels may manifest as localized texture disorder, abnormal fluctuations in feature values, etc., and it is difficult to accurately capture these subtle changes using only a single threshold. Furthermore, edge quality is a crucial indicator affecting the installation accuracy and appearance integrity of decorative panels. Existing detection methods for edge images primarily focus on simple geometric parameter measurements, lacking a comprehensive analysis of the correlation between edge defects and surface quality, resulting in insufficient comprehensiveness of the detection results.

[0005] During the continuous operation of a production line, the stability of the detection system directly affects the reliability of the detection results. When production conditions change, such as equipment wear and tear or raw material replacement, the original detection parameters may no longer be applicable. If the detection thresholds and parameters are not adjusted in a timely manner, it can lead to a large number of defective products entering the market or misjudging qualified products, increasing production costs and wasting resources. Traditional parameter adjustment methods mostly rely on manual periodic calibration, which is not only slow in response but also difficult to adjust accurately based on real-time production data, failing to meet the requirements of efficient production.

[0006] Furthermore, existing machine vision inspection methods have shortcomings in predicting and tracing quality fluctuations. When a defect is detected, it is often only possible to judge the current product, rather than analyzing the trends in texture feature changes in historical data to identify potential risks of quality fluctuations in advance, which is detrimental to proactive quality control on the production line. At the same time, for detected abnormal data, there is a lack of effective quantitative indicators to assess its severity, making it difficult to determine the priority of quality problem handling and affecting the timeliness and targeting of production adjustments. Summary of the Invention

[0007] The purpose of this invention is to provide a rapid inspection method for the production quality of building decorative panels based on machine vision, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, this invention provides a rapid inspection method for the production quality of architectural decorative panels based on machine vision, the method comprising the following steps:

[0009] The surface and edge images of each visual detection unit at each acquisition time in each detection cycle, as well as the sequence of each texture feature of each visual detection unit in each detection cycle, are acquired.

[0010] Based on the deviation between outliers in the texture feature sequence, the gradient of the texture feature sequence, and the deviation between the mean values ​​of each local texture feature sequence, the feature deviation index of each outlier in each texture feature sequence of each visual detection unit in each detection cycle is obtained.

[0011] The average deviation index of each visual detection unit at each quality fluctuation acquisition time in each detection cycle is obtained based on the distribution of the feature deviation index.

[0012] Based on the similarity and average deviation index between local texture feature data sequences, the quality fluctuation degree of each visual detection unit at each quality fluctuation acquisition time in each detection cycle is obtained.

[0013] The surface quality coefficient of each visual inspection unit in each inspection cycle is obtained based on the degree of quality fluctuation.

[0014] The edge quality coefficients of each visual detection unit in each detection cycle are obtained based on surface and edge images;

[0015] The adjustment values ​​of the detection parameters of each visual inspection unit in each inspection cycle are obtained based on the surface quality coefficient and the edge quality coefficient. The adjustment values ​​of the detection parameters of each visual inspection unit in each inspection cycle are used as the values ​​of the detection parameters of each visual inspection unit when performing defect identification by template matching in the next adjacent cycle of each inspection cycle, and the detection threshold is corrected.

[0016] Preferably, the surface image and edge image include:

[0017] For each visual inspection unit, the surface image at each acquisition time is acquired by a linear scan camera as the first surface image at each acquisition time, and the edge image at each acquisition time is acquired with the assistance of a ring light source as the first edge image at each acquisition time; the surface image and edge image at each acquisition time are acquired by an area scan camera as the second surface image and second edge image at each acquisition time.

[0018] Preferably, the method for obtaining the feature deviation index is as follows:

[0019] For each texture feature sequence of each visual detection unit in each detection period, the gray-scale threshold segmentation algorithm is used to obtain all the bright and dark areas in each texture feature sequence of each detection period as each anomaly point; for each anomaly point, the absolute value of the difference between the texture feature data of the anomaly point and the texture feature data of the previous adjacent acquisition time is calculated as the forward deviation index of each anomaly point in each texture feature sequence of each visual detection unit in each detection period.

[0020] The clustering index of each outlier is obtained based on the gradient of the texture feature data sequence.

[0021] For each abnormal point in each texture feature sequence of each visual detection unit in each detection cycle, a window of a preset size is constructed with the abnormal point as the center as the local window of each abnormal point. The mean of all texture feature data before the abnormal point in the local window is calculated, and the absolute value of the difference between the mean of all texture feature data after the abnormal point in the local window is used as the deviation of the mean before and after each abnormal point.

[0022] The feature deviation index is calculated as a weighted combination of forward deviation index, clustering index, and deviation from mean, with each weight being a preset adjustment coefficient.

[0023] Preferably, the method for obtaining the clustering index is as follows:

[0024] For each texture feature sequence of each visual detection unit in each detection cycle, the maximum value of the gradient magnitude of all elements in the texture feature sequence is calculated as the maximum gradient value of the texture feature sequence.

[0025] For each outlier in each texture feature sequence of each visual detection unit in each detection period, the absolute value of the difference between the texture feature data of the outlier and the maximum gradient value is calculated as the clustering index of each outlier.

[0026] Preferably, the method for obtaining the average deviation index is as follows:

[0027] For each outlier in each texture feature sequence of each visual detection unit in each detection period, when the texture feature data in all other types of texture feature sequences at the acquisition time of the outlier are all outliers, the acquisition time is taken as the quality fluctuation acquisition time. The mean of the feature deviation index of the outliers of all types of texture feature sequences at the quality fluctuation acquisition time of each detection period is calculated and used as the average deviation index of each visual detection unit at each quality fluctuation acquisition time of each detection period.

[0028] Preferably, the method for obtaining the degree of quality fluctuation is as follows:

[0029] For each quality fluctuation acquisition time in each detection cycle, the K-means clustering algorithm is used to cluster the average deviation index of all visual detection units to obtain each cluster; the cluster in which the average deviation index of each visual detection unit is located is taken as the target cluster of each visual detection unit.

[0030] For each visual detection unit, in each texture feature sequence of each detection cycle, at each quality fluctuation acquisition time, the sequence of all texture feature data in the local window of the texture feature data is taken as the local window sequence at each quality fluctuation acquisition time.

[0031] For each visual detection unit at each quality fluctuation acquisition time in each detection cycle, the mean similarity between the local window sequence of each texture feature sequence and the local window sequences of the same type of texture feature sequences of all other visual detection units in the target cluster is calculated as the feature trend difference of each texture feature sequence of each visual detection unit at each quality fluctuation acquisition time in each detection cycle. The mean of the feature trend differences of all types of texture feature sequences of each visual detection unit at each quality fluctuation acquisition time in each detection cycle is used as the feature dissimilarity index of each visual detection unit at each quality fluctuation acquisition time in each detection cycle.

[0032] The quality fluctuation level is calculated by a weighted combination of the average deviation index, the number of target cluster elements, and the feature dissimilarity index, with each weight being a preset adjustment factor.

[0033] Preferably, the method for obtaining the surface quality coefficient is as follows:

[0034] The quality fluctuation degree of each visual detection unit at all quality fluctuation acquisition times in each detection cycle is used as the input of the Otsu threshold segmentation algorithm, and the optimal segmentation threshold is output. Quality fluctuation acquisition times with a quality fluctuation degree greater than or equal to the optimal segmentation threshold are regarded as severe fluctuation acquisition times, and quality fluctuation acquisition times with a quality fluctuation degree less than the optimal segmentation threshold are regarded as stable fluctuation acquisition times.

[0035] For each visual inspection unit in each inspection cycle, the absolute value of the difference between the mean of the quality fluctuation degree at all severe fluctuation acquisition times and the mean of the quality fluctuation degree at all stable fluctuation acquisition times is calculated as the quality difference index of each visual inspection unit in each inspection cycle.

[0036] For each visual detection unit in each detection cycle, the ratio of the number of stable fluctuation acquisition moments to the total number of acquisition moments in the detection cycle is calculated as the proportion of stable duration for each visual detection unit in each detection cycle.

[0037] For each visual inspection unit in each inspection cycle, the product of the mean of the quality fluctuation degree at all severe fluctuation acquisition times and the quality difference index is calculated, and the ratio of the product to the proportion of the stable duration is used as the surface quality coefficient of each visual inspection unit in each inspection cycle.

[0038] Preferably, the method for obtaining the edge quality coefficient is as follows:

[0039] For each visual detection unit in each detection cycle, the average of the first surface image and the second surface image at each acquisition time is calculated as the initial surface feature of each visual detection unit in each detection cycle at each acquisition time, and the average of the first edge image and the second edge image at each acquisition time is calculated as the initial edge feature of each visual detection unit in each detection cycle at each acquisition time.

[0040] The edge quality coefficient is calculated by weighting the initial edge feature change and position difference at adjacent severe fluctuation acquisition times, with each weight determined based on the number of severe fluctuation acquisition times within the detection period.

[0041] Preferably, the method for obtaining the adjustment value of the detection parameter is as follows:

[0042] The adjustment value of the detection parameter is calculated by a weighted combination of a preset initial value, a surface quality coefficient, and an edge quality coefficient, with each weight determined by a normalization function.

[0043] Preferably, the method for determining the preset size of the local window is as follows:

[0044] For each texture feature sequence of each visual detection unit in each detection cycle, calculate the product of the number of all acquisition times in the detection cycle and the preset scaling factor, and use the calculation result as the preset size of the local window.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] A rapid quality inspection method for architectural decorative panels based on machine vision provides a more detailed analytical basis for quality inspection during the production process by collecting visual data from multiple dimensions. This method acquires surface images, edge images, and various texture feature sequences from each visual inspection unit at different inspection cycles, capturing quality information of the decorative panels from multiple levels and avoiding the bias that may result from a single data source. This comprehensive data acquisition approach can more realistically reflect the quality status of the decorative panels during the production process, allowing subsequent analysis and judgment to be based on richer information.

[0047] In outlier analysis, this method comprehensively considers the deviation between outliers in the texture feature sequence, the gradient of the sequence, and the deviation of the local mean to calculate a feature deviation index. This multi-factor analysis approach overcomes the limitations of traditional methods that rely solely on a single threshold to determine anomalies, enabling more accurate identification of outliers that truly affect quality and reducing misjudgments caused by random fluctuations. Furthermore, by analyzing the distribution of the feature deviation index to obtain the average deviation index at the time of quality fluctuation acquisition, the impact of outliers is further quantified, providing reliable basic data for subsequent quality fluctuation assessment.

[0048] To assess the degree of quality fluctuation, this method combines the similarity of local texture feature sequences and the average deviation index, enabling a more comprehensive reflection of quality stability at different sampling times. This assessment approach not only focuses on the impact of individual outliers but also considers the overall trend of texture feature changes, making the description of quality fluctuation more closely aligned with the actual quality variation patterns in production. The resulting surface quality coefficient accurately reflects the quality state of the decorative panel surface, providing an effective quantitative indicator for subsequent quality assessment.

[0049] In edge quality inspection, edge quality coefficients are obtained by combining surface and edge images, integrating surface and edge quality into a unified analytical framework to achieve a comprehensive assessment of decorative panel quality. This integrated assessment method avoids potential oversights that may occur when assessing surface or edge quality separately, ensuring the completeness of quality judgment. Furthermore, by combining surface and edge quality coefficients to determine adjustment values ​​for detection parameters, the detection system can optimize parameters in real time based on actual quality conditions, improving the adaptability of the inspection.

[0050] A key feature of this method is its dynamic adjustment mechanism for detection parameters. Based on the quality analysis results of the current cycle, the detection parameters for the next cycle are automatically adjusted, achieving self-optimization of the detection system. This dynamic adjustment can quickly respond to various changes in the production process, such as changes in raw material characteristics and fluctuations in equipment operating status, ensuring that the detection threshold remains within a reasonable range and reducing missed and false detections caused by fixed parameters. Furthermore, this parameter adjustment method based on actual data requires no manual intervention, reducing reliance on operator experience and making the detection process more automated and intelligent.

[0051] In practical production applications, this method can generate quality assessment results in real time and quickly feed them back to the parameter adjustments of the testing system, forming a closed-loop quality control process. This process not only improves the efficiency of quality inspection but also promptly identifies potential quality issues in the production process, facilitating timely adjustments by the production department and reducing the production of defective products. Simultaneously, through continuous analysis of texture feature sequences, a large amount of quality data can be accumulated, providing valuable reference information for improving production processes and fundamentally enhancing the production quality of architectural decorative panels. Attached Figure Description

[0052] Figure 1 This is a schematic diagram illustrating the working principle of the rapid quality inspection method for building decorative panels based on machine vision described in this invention.

[0053] Figure 2 A flowchart for calculating the characteristic deviation index;

[0054] Figure 3 A flowchart for calculating clustering indices;

[0055] Figure 4 A flowchart for calculating the degree of quality fluctuation. Detailed Implementation

[0056] 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.

[0057] Please see Figures 1-4 This invention provides a rapid inspection method for the production quality of building decorative panels based on machine vision. The method will be described in detail below with reference to the specific process.

[0058] Step 1: Acquire surface and edge images of each visual inspection unit at each acquisition moment in each inspection cycle, as well as each texture feature sequence of each visual inspection unit in each inspection cycle. Visual inspection units deployed at different locations on the production line acquire images of the building decorative panels according to a preset inspection cycle. Each inspection cycle contains multiple consecutive acquisition moments. At each acquisition moment, surface images, edge images, and multiple texture feature sequences are acquired simultaneously. These texture feature sequences include, but are not limited to, gray-level co-occurrence matrix features, oriented gradient histogram features, and other feature data that can reflect changes in the surface texture of the decorative panels.

[0059] Step 2: Based on the deviations between outliers in the texture feature sequence, the gradient of the texture feature sequence, and the deviations between the mean values ​​of the texture feature sequences in each locality, obtain the feature deviation index of each outlier in each texture feature sequence for each visual detection unit in each detection cycle. For each texture feature sequence, first identify the outliers in the sequence, and then quantify and calculate the feature deviation index corresponding to each outlier from three dimensions: the degree of mutual deviation between outliers, the gradient change trend of the overall sequence, and the difference in the mean value of texture features within a local range. This index is used to measure the degree of deviation of the outlier in the texture feature sequence.

[0060] Step 3: Based on the distribution of the feature deviation index, obtain the average deviation index of each visual detection unit at each quality fluctuation acquisition time in each detection cycle. Analyze the distribution of the feature deviation index at different acquisition times, screen out acquisition times with quality fluctuations, and calculate the average of the feature deviation indices of all relevant anomalies at these times as the average deviation index to comprehensively reflect the overall deviation level at that time.

[0061] Step 4: Based on the similarity and average deviation index between local texture feature data sequences, obtain the quality fluctuation degree of each visual detection unit at each quality fluctuation acquisition time in each detection cycle. By comparing the similarity of local texture feature sequences and combining the average deviation index, the fluctuation intensity at each quality fluctuation acquisition time is quantified to obtain the quality fluctuation degree. This index can reflect the quality stability of the decorative panel at that time.

[0062] Step 5: Obtain the surface quality coefficient of each visual inspection unit in each inspection cycle based on the degree of quality fluctuation. By comprehensively analyzing the degree of quality fluctuation at all acquisition times within the inspection cycle, a specific algorithm is used to obtain the surface quality coefficient, which characterizes the overall surface quality of the building decorative panel.

[0063] Step 6: Obtain the edge quality coefficient of each visual detection unit in each detection cycle based on the surface image and edge image. Perform image processing and feature extraction on the acquired surface image and edge image, analyze the integrity, smoothness and other features of the edges, and calculate the edge quality coefficient to reflect the quality of the edge part of the decorative panel.

[0064] Step 7: Based on the surface quality coefficient and edge quality coefficient, obtain the adjustment values ​​of the detection parameters for each visual inspection unit in each inspection cycle. Use these adjusted values ​​as the detection parameter values ​​for each visual inspection unit when performing template matching for defect identification in the next adjacent cycle, thus correcting the detection threshold. By fusing the surface quality coefficient and edge quality coefficient, the adjusted detection parameter values ​​are calculated and applied to the template matching defect identification process in the next inspection cycle, achieving dynamic correction of the detection threshold to improve detection accuracy and adaptability.

[0065] Example 1:

[0066] For each vision inspection unit, a collaborative approach using line scan cameras and area scan cameras is employed during the acquisition of surface and edge images. The line scan camera, with its line-by-line scanning capability, can acquire surface images at high speeds on a continuously transporting production line of architectural decorative panels; these images are defined as the first surface image. To enhance the imaging effect in edge areas, a ring light source is used for auxiliary illumination when acquiring edge images using the line scan camera. The ring light source provides uniform illumination to the edges of the decorative panels from multiple angles, reducing the impact of shadows and reflections on image quality. The edge images acquired by the line scan camera at each acquisition moment with the assistance of the ring light source are defined as the first edge image.

[0067] An area scan camera can capture images of a large area at once. At each acquisition moment, the area scan camera simultaneously captures images of the surface and edges of the decorative panel, obtaining a second surface image and a second edge image, respectively. The first surface image and the second surface image differ in their imaging principles. The first surface image from a line scan camera can show the continuous texture features of the decorative panel surface along the direction of movement, while the second surface image from an area scan camera can present the overall regional features of the decorative panel surface at a certain moment. Combining the two can comprehensively reflect the details and overall condition of the surface.

[0068] The first and second edge images are complementary. The first edge image acquired by the line scan camera performs better in terms of the continuity of edge lines, making it suitable for analyzing the straightness and continuity of edges; while the second edge image acquired by the area scan camera can more clearly show the junction between the edge and the surface, making it easier to observe the integrity and transition state of the edge.

[0069] In practical applications, the installation positions of the line scan camera and the area scan camera need to be precisely calibrated to ensure that the images acquired by both accurately correspond to the same area of ​​the decorative panel. During calibration, the height, angle, and focal length of the cameras are adjusted to ensure that the image areas captured by the two cameras at the same acquisition moment completely overlap, avoiding image misalignment caused by differences in field of view. Simultaneously, the acquisition frequency of both cameras needs to match the operating speed of the production line. When the production line speed changes, the acquisition frequency is adjusted synchronously to ensure that clear, blur-free image data is acquired at every acquisition moment.

[0070] The scanning frequency of the line scan camera is set according to the width of the decorative panel and the transmission speed to ensure that the entire width direction can be scanned within a unit time. The shooting frequency of the area scan camera is determined according to the detection accuracy requirements, avoiding data redundancy caused by excessive frequency while meeting the detection needs. The brightness of the ring light source also needs to be adjusted according to the surface reflectivity of the decorative panel. For decorative panels with high gloss, the brightness of the light source is appropriately reduced to reduce reflection; for decorative panels with darker surfaces, the brightness of the light source is increased to improve image contrast.

[0071] Through the coordinated operation of linear and area scan cameras, and with the assistance of a ring light source, the acquired first surface image, second surface image, first edge image, and second edge image can comprehensively and accurately reflect the surface and edge features of the building decorative panel at each acquisition time, providing rich and reliable raw data for subsequent quality analysis and adjustment of detection parameters.

[0072] Example 2:

[0073] For each texture feature sequence in each detection cycle of each visual detection unit, obtaining the feature deviation index requires a series of steps. First, a grayscale thresholding algorithm is used to process the texture feature sequence. This algorithm divides the parts of the sequence whose grayscale values ​​exceed the normal range into highlight areas and dark areas by setting a grayscale threshold. These areas are identified as outliers in the texture feature sequence. The grayscale threshold needs to be set in conjunction with the normal value range of the texture features and determined by analyzing historical normal texture data to ensure accurate differentiation between normal areas and outliers.

[0074] For each outlier, the absolute value of the difference between its texture feature data and the texture feature data at the previous adjacent acquisition time is calculated. This value is the forward deviation index. The forward deviation index reflects the magnitude of the change in texture features of the outlier over time compared to the previous time. If there is no data at the previous adjacent acquisition time, the texture feature data at the first acquisition time in this detection period is used as the reference benchmark.

[0075] When obtaining the clustering index, the gradient magnitude of all elements in the texture feature sequence is first calculated. The gradient magnitude is obtained by the difference between the texture feature data at two adjacent acquisition times, and the maximum value among all gradient magnitudes is taken as the maximum gradient value of the texture feature sequence. For each outlier, the absolute value of the difference between its texture feature data and the maximum gradient value is calculated. This value is the clustering index, which reflects the deviation of the outlier from the maximum magnitude of gradient change in the entire sequence.

[0076] When constructing a local window, each outlier is centered on a pre-defined window size, and all texture feature data within the window's range are included in the analysis. The mean of all texture feature data before the outlier is calculated, followed by the mean of all texture feature data after the outlier. The absolute value of the difference between these two means is the mean deviation. This mean deviation reflects the overall difference in texture features between the two segments within the local area containing the outlier. If the outlier is located at the beginning of the window, only the absolute value of the difference between the mean of texture feature data after the outlier and the texture feature data at the outlier is calculated; if the outlier is located at the end of the window, only the absolute value of the difference between the mean of texture feature data before the outlier and the texture feature data at the outlier is calculated.

[0077] The feature deviation index is obtained by weighting forward deviation index, clustering index, and deviation from the mean, with each weight being a preset adjustment coefficient. The value of the adjustment coefficient needs to be determined according to the importance of different texture features in quality inspection. For example, for texture features related to surface flatness, the weight of the forward deviation index can be appropriately increased; for features related to texture consistency, the weight of the deviation from the mean can be increased accordingly. In practical applications, the adjustment coefficient can be set after analyzing the inspection data of various decorative panel samples and can be dynamically adjusted according to quality feedback during the production process to ensure that the feature deviation index can accurately quantify the degree of deviation of outliers, providing a reliable basis for subsequent quality fluctuation analysis.

[0078] Throughout the process, for each texture feature sequence, the feature deviation index of outliers needs to be calculated independently. The processing of different texture feature sequences is independent of each other, ensuring that the outliers of each feature can be accurately captured and quantified.

[0079] Example 3:

[0080] For each outlier in each texture feature sequence of each visual detection unit in each detection cycle, it is determined whether the acquisition time of that outlier is a quality fluctuation acquisition time. When all texture feature data in all types of texture feature sequences at a certain acquisition time are outliers, that acquisition time is determined to be a quality fluctuation acquisition time. After determining the quality fluctuation acquisition time, the mean of the feature deviation index of the outliers in all types of texture feature sequences at that time is calculated, and this is used as the average deviation index of each visual detection unit at that quality fluctuation acquisition time in that detection cycle.

[0081] To calculate the degree of quality fluctuation, the K-means clustering algorithm is first used to cluster the average deviation index of all visual inspection units at each quality fluctuation acquisition time of each detection cycle. During the clustering process, based on the distribution characteristics of the average deviation index, average deviation indices with similar values ​​are grouped into the same cluster, resulting in several clusters. The cluster containing the average deviation index of each visual inspection unit is determined as its target cluster, reflecting the similarity of that visual inspection unit to other units in terms of the degree of quality fluctuation.

[0082] Next, for each visual detection unit, at each quality fluctuation acquisition time in each texture feature sequence of each detection cycle, a local window of a preset size is constructed centered on the anomaly point. All texture feature data in the local window are arranged in chronological order to form a sequence, which serves as the local window sequence at that quality fluctuation acquisition time. The size of the local window is determined comprehensively based on the change frequency of the texture feature sequence and the length of the detection cycle to ensure that it can cover sufficient historical and subsequent data to reflect the texture feature change trend before and after the anomaly point.

[0083] For each visual detection unit at each quality fluctuation acquisition time in each detection cycle, the similarity between the local window sequence of each texture feature sequence and the local window sequences of the same type of texture feature sequences of all other visual detection units in the target cluster is calculated. Similarity calculation is achieved by comparing the overall trend and numerical distribution of the two sequences; the closer the values ​​are, the more consistent the trend, and the higher the similarity. The average of all similarity values ​​is taken as the feature trend difference of that texture feature sequence of that visual detection unit at the corresponding detection cycle and quality fluctuation acquisition time. The smaller the feature trend difference, the more consistent the texture feature change of that visual detection unit is with other units in the target cluster; conversely, a larger difference indicates a more significant difference.

[0084] The feature dissimilarity index is obtained by averaging the feature trend differences of all types of texture feature sequences of the visual detection unit during the detection period and at the quality fluctuation acquisition time. The feature dissimilarity index comprehensively reflects the overall degree of difference between the visual detection unit and other units within the target cluster in various texture features.

[0085] Finally, the degree of quality fluctuation is calculated by a weighted combination of the average deviation index, the number of elements in the target cluster, and the feature dissimilarity index, as shown in the following formula:

[0086] Z = a × P + b × Q + c × R

[0087] Where Z represents the degree of quality fluctuation, P represents the average deviation index, Q represents the number of target cluster elements, R represents the feature dissimilarity index, and a, b, and c are the weights of the average deviation index, the number of target cluster elements, and the feature dissimilarity index, respectively, and are all preset adjustment factors.

[0088] In practical applications, the value of the adjustment factor is determined based on the detection scenario and the quality characteristics of the decorative panel. When the average deviation index has a significant impact on quality fluctuation, the value of 'a' is increased accordingly; if the number of target cluster elements is large, indicating that the fluctuation pattern is universal, the value of 'b' can be appropriately increased; and when the characteristic dissimilarity index better reflects the impact of individual differences on quality, the value of 'c' needs to be adjusted. Through this weighted combination method, the degree of quality fluctuation can comprehensively integrate information from multiple dimensions, reflecting both the absolute level of quality fluctuation at that moment and its relative position and characteristic differences within the group, providing a quantitative basis for subsequent surface quality coefficient calculation.

[0089] Throughout the process, the degree of quality fluctuation needs to be calculated independently for each inspection cycle and each time quality fluctuation is collected, ensuring that dynamic changes in quality during production can be captured in real time. The degree of quality fluctuation between different vision inspection units can be directly compared, facilitating the identification of abnormal areas or equipment on the production line and providing targeted reference information for quality control. Simultaneously, the division of target clusters and the calculation of feature similarity are based on actually collected data, avoiding interference from subjective factors and making the calculation results of the degree of quality fluctuation more objective and reliable.

[0090] Example 4:

[0091] When acquiring the surface quality coefficient, the degree of quality fluctuation for each visual inspection unit at all quality fluctuation acquisition moments in each inspection cycle is used as input data for the Otsu threshold segmentation algorithm. The Otsu threshold segmentation algorithm automatically determines an optimal segmentation threshold by analyzing the grayscale distribution characteristics of the quality fluctuation degree. This threshold divides the quality fluctuation degree into two different categories. Quality fluctuation acquisition moments with a quality fluctuation degree greater than or equal to the optimal segmentation threshold are classified as severe fluctuation acquisition moments; quality fluctuation acquisition moments with a quality fluctuation degree less than the optimal segmentation threshold are classified as stable fluctuation acquisition moments.

[0092] For each visual inspection unit in each inspection cycle, it is necessary to calculate the mean of the quality fluctuation degree at the severe fluctuation acquisition time and the steady fluctuation acquisition time, respectively. Specifically, the quality fluctuation degree of the visual inspection unit at all severe fluctuation acquisition times within the current inspection cycle is collected, and the arithmetic mean of these values ​​is calculated to obtain the severe fluctuation mean. Similarly, the quality fluctuation degree at all steady fluctuation acquisition times is collected, and its arithmetic mean is calculated to obtain the steady fluctuation mean. Then, the absolute value of the difference between the severe fluctuation mean and the steady fluctuation mean is calculated, and this absolute value is used as the quality difference index of the visual inspection unit in the current inspection cycle. The magnitude of the quality difference index reflects the gap between severe fluctuation and steady fluctuation; the larger the difference index, the more significant the difference in the quality fluctuation degree between the two fluctuation states.

[0093] Simultaneously, the number of stable fluctuation acquisition moments of the visual inspection unit within the current inspection cycle, and the total number of acquisition moments within the entire inspection cycle, are counted. The ratio of the number of stable fluctuation acquisition moments to the total number is calculated to obtain the stable duration percentage. The value of the stable duration percentage ranges from 0 to 1. The larger the ratio, the higher the proportion of time the decorative panel is in a stable fluctuation state throughout the entire inspection cycle.

[0094] Finally, the mean of severe fluctuations is multiplied by the quality difference index to obtain a product. This product is then divided by the percentage of stable time, and the resulting quotient is the surface quality coefficient of the visual inspection unit in the current inspection cycle.

[0095] When obtaining the edge quality coefficient, the image data of each visual detection unit at each acquisition time in each detection cycle are first processed. For each acquisition time, the average of the first surface image acquired by the line scan camera and the second surface image acquired by the area scan camera is calculated to obtain the initial surface features at that acquisition time. Similarly, the average of the first edge image acquired by the line scan camera under the assistance of a ring light source and the second edge image acquired by the area scan camera is calculated to obtain the initial edge features at that acquisition time. Both the initial surface features and the initial edge features are presented in the form of image pixel values, which can reflect the basic features of the decorative panel surface and edges at that time.

[0096] For each visual detection unit in each detection cycle, severely fluctuating acquisition times are selected from all acquisition times and sorted chronologically. Then, two adjacent severely fluctuating acquisition times are selected sequentially, and their initial edge feature changes and positional differences are calculated. The initial edge feature change refers to the absolute value of the difference between the initial edge features of the later severely fluctuating acquisition time and the initial edge features of the earlier severely fluctuating acquisition time; the positional difference refers to the difference between the positional number of the later severely fluctuating acquisition time within the detection cycle and the positional number of the earlier severely fluctuating acquisition time.

[0097] The initial edge feature change and the positional difference are weighted and combined to obtain the edge quality sub-coefficients corresponding to adjacent severe fluctuation acquisition times. For all adjacent severe fluctuation acquisition times within the same detection period, the edge quality sub-coefficients are calculated in the above manner, and then the average of all edge quality sub-coefficients is taken as the edge quality coefficient of the visual detection unit in the current detection period. The weights used in the weighted combination are determined according to the number of severe fluctuation acquisition times within the detection period. When the number of severe fluctuation acquisition times is large, the weight of the positional difference will be adjusted accordingly to balance the impact of edge feature changes at different intervals on the overall edge quality.

[0098] Example 5:

[0099] The adjustment values ​​for the testing parameters need to be obtained by combining the preset initial values, surface quality coefficient, and edge quality coefficient. The preset initial values ​​are determined based on the type and material of the architectural decorative panel and the basic testing standards. Different types of decorative panels correspond to different preset initial values. For example, the preset initial values ​​for wood decorative panels and stone decorative panels will be set differently based on their respective surface characteristics and common defect types. The surface quality coefficient and edge quality coefficient are derived from the quality analysis results of the current testing cycle, reflecting the quality status of the decorative panel's surface and edges, respectively.

[0100] When calculating the adjustment values ​​of the detection parameters, a weighted combination of the preset initial value, surface quality coefficient, and edge quality coefficient is required. Each weight is determined by a normalization function, which processes the numerical ranges of the surface quality coefficient and edge quality coefficient, converting them into corresponding weight ratios. Specifically, the normalization function first standardizes the surface quality coefficient and edge quality coefficient to eliminate the influence of different dimensions, and then assigns weights based on their relative importance in quality inspection. The weight of the preset initial value is 1 minus the sum of the weights of the surface quality coefficient and edge quality coefficient, ensuring that the total weight of the three is 1. Through this weighted combination, the resulting adjustment value of the detection parameters comprehensively reflects the baseline requirements and the current quality status. This value serves as the detection parameter value for each visual inspection unit when using template matching for defect identification in the next adjacent cycle, thereby correcting the detection threshold and enabling it to adapt to quality fluctuations during the production process.

[0101] The preset size of the local window needs to be determined separately for each texture feature sequence of each visual inspection unit in each inspection cycle. During the determination process, firstly, the number of all acquisition moments within the inspection cycle is counted. The number of acquisition moments is set according to the production line's operating speed and inspection accuracy requirements; the number of acquisition moments increases accordingly when the operating speed is high or the inspection accuracy requirements are high. Then, the number of acquisition moments is multiplied by a preset scaling factor; the product is the preset size of the local window for that texture feature sequence.

[0102] The value of the preset scaling factor needs to consider the changing characteristics of the texture feature sequence. For texture features that change frequently, such as the density variation of surface texture, the preset scaling factor can be appropriately increased to allow the local window to cover more acquisition moments and capture a more comprehensive texture change trend. For texture features that change relatively smoothly, such as the smoothness of edges, the preset scaling factor can be appropriately decreased to improve the targeting of local analysis. The specific value of the preset scaling factor can be determined based on historical detection data and the production process of the decorative panel. The preset scaling factor can be different in different detection cycles and different texture feature sequences to adapt to diverse detection needs.

[0103] In practical applications, the preset size of the local window is dynamically adjusted according to the detection cycle and the texture feature sequence. For example, in a certain detection cycle, if the number of acquisition times for a certain texture feature sequence is 100 and the preset ratio coefficient is 0.2, then the preset size of the local window for that texture feature sequence is 20, meaning that each local window contains texture feature data from 20 acquisition times. The local window size determined in this way matches the variation pattern of the texture features, providing a suitable analysis range for steps such as calculating the deviation of the mean before and after anomalies. This ensures that the texture feature data within the local window accurately reflects the local environment where the anomaly is located, thereby improving the reliability of calculation results such as the feature deviation index.

[0104] The adjustment of the detection parameters and the determination of the preset size of the local window are independent of each other, but both are based on the actual data of the current detection cycle. This ensures that the entire detection method can be dynamically adjusted according to the actual production situation, adapting to different production conditions and quality statuses, and providing flexible and reliable technical support for the rapid detection of the production quality of building decorative panels.

[0105] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A rapid quality inspection method for architectural decorative panels based on machine vision, characterized in that, The method includes the following steps: The surface and edge images of each visual detection unit at each acquisition time in each detection cycle, as well as the sequence of each texture feature of each visual detection unit in each detection cycle, are acquired. Based on the deviation between outliers in the texture feature sequence, the gradient of the texture feature sequence, and the deviation between the mean values ​​of each local texture feature sequence, the feature deviation index of each outlier in each texture feature sequence of each visual detection unit in each detection cycle is obtained. The average deviation index of each visual detection unit at each quality fluctuation acquisition time in each detection cycle is obtained based on the distribution of the feature deviation index. Based on the similarity and average deviation index between local texture feature data sequences, the quality fluctuation degree of each visual detection unit at each quality fluctuation acquisition time in each detection cycle is obtained. The surface quality coefficient of each visual inspection unit in each inspection cycle is obtained based on the degree of quality fluctuation. The edge quality coefficients of each visual detection unit in each detection cycle are obtained based on surface and edge images; The adjustment values ​​of the detection parameters of each visual inspection unit in each inspection cycle are obtained based on the surface quality coefficient and the edge quality coefficient. The adjustment values ​​of the detection parameters of each visual inspection unit in each inspection cycle are used as the values ​​of the detection parameters of each visual inspection unit when performing defect identification by template matching in the next adjacent cycle of each inspection cycle, and the detection threshold is corrected.

2. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 1, characterized in that, The surface image and edge image include: For each visual inspection unit, the surface image at each acquisition time is acquired by a linear scan camera as the first surface image at each acquisition time, and the edge image at each acquisition time is acquired with the assistance of a ring light source as the first edge image at each acquisition time; the surface image and edge image at each acquisition time are acquired by an area scan camera as the second surface image and second edge image at each acquisition time.

3. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 1, characterized in that, The method for obtaining the feature deviation index is as follows: For each texture feature sequence of each visual detection unit in each detection period, the gray-scale threshold segmentation algorithm is used to obtain all the bright and dark areas in each texture feature sequence of each detection period as each anomaly point; for each anomaly point, the absolute value of the difference between the texture feature data of the anomaly point and the texture feature data of the previous adjacent acquisition time is calculated as the forward deviation index of each anomaly point in each texture feature sequence of each visual detection unit in each detection period. The clustering index of each outlier is obtained based on the gradient of the texture feature data sequence. For each abnormal point in each texture feature sequence of each visual detection unit in each detection cycle, a window of a preset size is constructed with the abnormal point as the center as the local window of each abnormal point. The mean of all texture feature data before the abnormal point in the local window is calculated, and the absolute value of the difference between the mean of all texture feature data after the abnormal point in the local window is used as the deviation of the mean before and after each abnormal point. The feature deviation index is calculated as a weighted combination of forward deviation index, clustering index, and deviation from mean, with each weight being a preset adjustment coefficient.

4. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 3, characterized in that, The method for obtaining the clustering index is as follows: For each texture feature sequence of each visual detection unit in each detection cycle, the maximum value of the gradient magnitude of all elements in the texture feature sequence is calculated as the maximum gradient value of the texture feature sequence. For each outlier in each texture feature sequence of each visual detection unit in each detection period, the absolute value of the difference between the texture feature data of the outlier and the maximum gradient is calculated as the clustering index of each outlier.

5. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 1, characterized in that, The method for obtaining the average deviation index is as follows: For each outlier in each texture feature sequence of each visual detection unit in each detection period, when the texture feature data in all other types of texture feature sequences at the acquisition time of the outlier are all outliers, the acquisition time is taken as the quality fluctuation acquisition time. The mean of the feature deviation index of the outliers of all types of texture feature sequences at the quality fluctuation acquisition time of each detection period is calculated and used as the average deviation index of each visual detection unit at each quality fluctuation acquisition time of each detection period.

6. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 3, characterized in that, The method for obtaining the degree of quality fluctuation is as follows: For each quality fluctuation acquisition time in each detection cycle, the K-means clustering algorithm is used to cluster the average deviation index of all visual detection units to obtain each cluster; the cluster in which the average deviation index of each visual detection unit is located is taken as the target cluster of each visual detection unit. For each visual detection unit, in each texture feature sequence of each detection cycle, at each quality fluctuation acquisition time, the sequence of all texture feature data in the local window of the texture feature data is taken as the local window sequence at each quality fluctuation acquisition time. For each visual detection unit at each quality fluctuation acquisition time in each detection cycle, the mean similarity between the local window sequence of each texture feature sequence and the local window sequences of the same type of texture feature sequences of all other visual detection units in the target cluster is calculated as the feature trend difference of each texture feature sequence of each visual detection unit at each quality fluctuation acquisition time in each detection cycle. The mean of the feature trend differences of all types of texture feature sequences of each visual detection unit at each quality fluctuation acquisition time in each detection cycle is used as the feature dissimilarity index of each visual detection unit at each quality fluctuation acquisition time in each detection cycle. The quality fluctuation level is calculated by a weighted combination of the average deviation index, the number of target cluster elements, and the feature dissimilarity index, with each weight being a preset adjustment factor.

7. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 2, characterized in that, The method for obtaining the surface quality coefficient is as follows: The quality fluctuation degree of each visual detection unit at all quality fluctuation acquisition times in each detection cycle is used as the input of the Otsu threshold segmentation algorithm, and the optimal segmentation threshold is output. Quality fluctuation acquisition times with a quality fluctuation degree greater than or equal to the optimal segmentation threshold are regarded as severe fluctuation acquisition times, and quality fluctuation acquisition times with a quality fluctuation degree less than the optimal segmentation threshold are regarded as stable fluctuation acquisition times. For each visual inspection unit in each inspection cycle, the absolute value of the difference between the mean of the quality fluctuation degree at all severe fluctuation acquisition times and the mean of the quality fluctuation degree at all stable fluctuation acquisition times is calculated as the quality difference index of each visual inspection unit in each inspection cycle. For each visual detection unit in each detection cycle, the ratio of the number of stable fluctuation acquisition moments to the total number of acquisition moments in the detection cycle is calculated as the proportion of stable duration for each visual detection unit in each detection cycle. For each visual inspection unit in each inspection cycle, the product of the mean of the quality fluctuation degree at all severe fluctuation acquisition times and the quality difference index is calculated, and the ratio of the product to the proportion of the stable duration is used as the surface quality coefficient of each visual inspection unit in each inspection cycle.

8. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 7, characterized in that, The method for obtaining the edge quality coefficient is as follows: For each visual detection unit in each detection cycle, the average of the first surface image and the second surface image at each acquisition time is calculated as the initial surface feature of each visual detection unit in each detection cycle at each acquisition time, and the average of the first edge image and the second edge image at each acquisition time is calculated as the initial edge feature of each visual detection unit in each detection cycle at each acquisition time. The edge quality coefficient is calculated by weighting the initial edge feature change and position difference at adjacent severe fluctuation acquisition times, with each weight determined based on the number of severe fluctuation acquisition times within the detection period.

9. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 1, characterized in that, The method for obtaining the adjustment value of the detection parameter is as follows: The adjustment value of the detection parameter is calculated by a weighted combination of a preset initial value, a surface quality coefficient, and an edge quality coefficient, with each weight determined by a normalization function.

10. The rapid quality inspection method for building decorative panels based on machine vision as described in claim 3, characterized in that, The method for determining the preset size of the local window is as follows: For each texture feature sequence of each visual detection unit in each detection cycle, calculate the product of the number of all acquisition times in the detection cycle and the preset scaling factor, and use the calculation result as the preset size of the local window.

Images