A Defect Detection Method and System for Rock Wool Curtain Wall Panels Based on Machine Vision

Abstract

The present invention relates to the technical field of wall panel defect detection, and particularly to a method and system for detecting defects of rock wool curtain wall panels based on machine vision. The method includes the following steps: performing multi-spectral image acquisition on the surface of the rock wool curtain wall panel to obtain a multi-spectral image set of the rock wool curtain wall panel; extracting features from the multi-spectral image set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel; performing geometric defect segmentation on the initial feature map of the rock wool curtain wall panel to obtain a geometric defect feature map of the rock wool curtain wall panel; performing material anomaly classification on the initial feature map of the rock wool curtain wall panel to obtain a material defect feature map of the rock wool curtain wall panel; performing feature alignment and fusion on the geometric defect feature map of the rock wool curtain wall panel and the material defect feature map of the rock wool curtain wall panel to obtain a comprehensive defect feature map of the rock wool curtain wall panel. The present invention not only improves the efficiency and accuracy of defect detection of rock wool curtain wall panels, but also reduces costs and resource waste.

Description

Technical Field

[0001] The present invention relates to the technical field of wall panel defect detection, and particularly to a method and system for detecting defects of rock wool curtain wall panels based on machine vision. Background Art

[0002] Although existing automatic detection devices for rock wool curtain wall panels can improve the detection efficiency to a certain extent, when classifying rock wool curtain wall panels, they often require a long transportation main line to cooperate with branch lines. This design not only increases the floor area of the equipment, but also raises the overall cost. For example, a typical rock wool curtain wall panel production line requires dozens of meters of transportation lines to ensure the smooth transportation of rock wool curtain wall panels from the detection point to the classification point. This not only occupies a large amount of factory space, but also increases the construction cost and maintenance cost. In addition, during the process of classifying and transporting rock wool curtain wall panels, the existing devices have limited ability to identify and classify rock wool curtain wall panels with different defect levels. This means that they cannot accurately distinguish between slightly defective and severely defective rock wool curtain wall panels, resulting in the inability to achieve refined management. For example, some devices can only perform simple classification of qualified and unqualified, and cannot further classify products into more detailed categories such as first-class products and second-class products. Due to the lack of classification ability, many repairable or reusable rock wool curtain wall panels are easily misclassified as waste products, leading to waste of resources. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide a method and system for detecting defects of rock wool curtain wall panels based on machine vision to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for detecting defects of rock wool curtain wall panels based on machine vision includes the following steps:

[0005] Step S1: Collect multi-spectral images of the surface of the rock wool curtain wall panel to obtain a multi-spectral image set of the rock wool curtain wall panel; extract features from the multi-spectral image set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel;

[0006] Step S2: Segment geometric defects of the initial feature map of the rock wool curtain wall panel to obtain a geometric defect feature map of the rock wool curtain wall panel; divide material anomalies of the initial feature map of the rock wool curtain wall panel to obtain a material defect feature map of the rock wool curtain wall panel; align and fuse the geometric defect feature map and the material defect feature map of the rock wool curtain wall panel to obtain a comprehensive defect feature map of the rock wool curtain wall panel;

[0007] Step S3: Calculate defect quantification indexes for the comprehensive defect feature map of the rock wool curtain wall panel to obtain a defect feature vector of the rock wool curtain wall panel; perform defect grading mapping on the defect feature vector of the rock wool curtain wall panel to obtain a defect label of the rock wool curtain wall panel;

[0008] Step S4: According to the rock wool curtain wall board defect labels, conduct intelligent classification and guidance for the rock wool curtain wall boards to obtain the classification planning path of the rock wool curtain wall boards; conduct collaborative control of the flexible conveying system for the classification planning path of the rock wool curtain wall boards to obtain the classification conveying plan of the rock wool curtain wall boards; position the rock wool curtain wall boards according to the classification conveying plan of the rock wool curtain wall boards to obtain the classification storage coordinates of the rock wool curtain wall boards;

[0009] Step S5: Conduct detection data association for the storage coordinates of the classified rock wool curtain wall boards to obtain the detection and traceability link of the rock wool curtain wall boards; conduct digital reconstruction of the production process based on the detection and traceability link to obtain the production quality management model; generate optimization suggestions for the production process of the rock wool curtain wall boards according to the production quality management model to obtain the production suggestion report of the rock wool curtain wall boards; apply the production suggestion report of the rock wool curtain wall boards to the rock wool curtain wall board production line, and execute Steps S1 - S3 for the newly produced rock wool curtain wall boards, thereby obtaining new rock wool curtain wall board defect labels.

[0010] Through multi - spectral image acquisition and feature extraction, the present invention can more meticulously identify the geometric and material defects on the surface of the rock wool curtain wall boards, thereby improving the accuracy of defect detection. Through defect quantification index calculation and defect grading mapping, the detection results are made more refined, capable of distinguishing defects of different severity levels, and improving the detection efficiency. Through the accurate defect grading ability, the rock wool curtain wall boards can be classified according to the severity of the defects, avoiding misclassifying repairable or reusable rock wool curtain wall boards as waste products and reducing resource waste. Through intelligent classification guidance and collaborative control of the flexible conveying system, the dependence on the long transportation main line and branch lines is reduced, thereby reducing the floor area and overall cost of the equipment. Through detection data association and digital reconstruction of the production process, a production quality management model can be generated, providing a basis for the optimization of the production process. In summary, the present invention not only improves the efficiency and accuracy of rock wool curtain wall board defect detection, but also reduces costs and resource waste.

[0011] Preferably, the present invention also provides a machine - vision - based rock wool curtain wall board defect detection system for executing the above - mentioned machine - vision - based rock wool curtain wall board defect detection method. The machine - vision - based rock wool curtain wall board defect detection system includes:

[0012] An image acquisition module, which is used to conduct multi - spectral image acquisition on the surface of the rock wool curtain wall boards to obtain a multi - spectral image set of the rock wool curtain wall boards; conduct feature extraction on the multi - spectral image set of the rock wool curtain wall boards to obtain an initial feature map of the rock wool curtain wall boards;

[0013] A defect extraction module is used to perform geometric defect segmentation on the initial feature map of the rock wool curtain wall board to obtain the geometric defect feature map of the rock wool curtain wall board; perform material anomaly classification on the initial feature map of the rock wool curtain wall board to obtain the material defect feature map of the rock wool curtain wall board; perform feature alignment and fusion on the geometric defect feature map of the rock wool curtain wall board and the material defect feature map of the rock wool curtain wall board to obtain the comprehensive defect feature map of the rock wool curtain wall board;

[0014] A defect classification module is used to calculate defect quantification indexes for the comprehensive defect feature map of the rock wool curtain wall board to obtain the defect feature vector of the rock wool curtain wall board; perform defect grading mapping on the defect feature vector of the rock wool curtain wall board to obtain the defect label of the rock wool curtain wall board;

[0015] A classification control module is used to perform intelligent classification guidance on the rock wool curtain wall board according to the defect label of the rock wool curtain wall board to obtain the classification planning path of the rock wool curtain wall board; perform collaborative control of the flexible conveying system on the classification planning path of the rock wool curtain wall board to obtain the classification conveying scheme of the rock wool curtain wall board; position the rock wool curtain wall board according to the classification conveying scheme of the rock wool curtain wall board to obtain the classification storage coordinates of the rock wool curtain wall board;

[0016] A traceability and optimization module is used to associate detection data with the storage coordinates of the classified rock wool curtain wall board to obtain the detection traceability link of the rock wool curtain wall board; perform digital reconstruction of the production process based on the detection traceability link to obtain the production quality management model; generate optimization suggestions for the production process of the rock wool curtain wall board according to the production quality management model to obtain the production suggestion report of the rock wool curtain wall board; apply the production suggestion report of the rock wool curtain wall board to the rock wool curtain wall board production line, and execute steps S1 - S3 for the newly produced rock wool curtain wall board to obtain the defect label of the new rock wool curtain wall board.

[0017] The present invention improves the automation level of the production line through automated image acquisition and feature extraction, and at the same time improves the accuracy of defect detection by using multi - spectral imaging technology. The system performs detailed segmentation and classification of the geometric and material defects of the rock wool curtain wall board, optimizes defect management, and improves the classification efficiency through intelligent classification guidance and collaborative control of the flexible conveying system. This not only reduces the risk of misclassifying repairable or recyclable rock wool curtain wall boards as waste, reduces resource waste, but also reduces production costs by reducing manual inspection and optimizing the classification process. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects, and advantages of the present invention will become more apparent by reading the following detailed description with reference to the accompanying drawings:

[0019] Figure 1 Shows the step - flow schematic diagram of a method for detecting defects of rock wool curtain wall boards based on machine vision in an embodiment.

[0020] Figure 2Shows a detailed step - by - step flowchart of step S25 of an embodiment.

[0021] Figure 3 Shows a detailed step - by - step flowchart of step S35 of an embodiment. Specific implementation manner

[0022] The following clearly and completely describes the technical method of the present invention with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present invention.

[0023] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, so repeated descriptions of them will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0024] It should be understood that although terms such as "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed associated items.

[0025] To achieve the above - mentioned purpose, please refer to Figures 1 to 3 , the present invention provides a method for detecting defects of rock wool curtain wall panels based on machine vision, including the following steps:

[0026] Step S1: Perform multi - spectral image acquisition on the surface of the rock wool curtain wall panel to obtain a multi - spectral image set of the rock wool curtain wall panel; extract features from the multi - spectral image set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel;

[0027] Step S2: Perform geometric defect segmentation on the initial feature map of the rock wool curtain wall panel to obtain a geometric defect feature map of the rock wool curtain wall panel; perform material anomaly division on the initial feature map of the rock wool curtain wall panel to obtain a material defect feature map of the rock wool curtain wall panel; perform feature alignment and fusion on the geometric defect feature map and the material defect feature map of the rock wool curtain wall panel to obtain a comprehensive defect feature map of the rock wool curtain wall panel;

[0028] Step S3: Calculate the defect quantification index for the comprehensive feature map of the rock wool curtain wall board to obtain the defect feature vector of the rock wool curtain wall board; perform defect grading mapping on the defect feature vector of the rock wool curtain wall board to obtain the defect label of the rock wool curtain wall board;

[0029] Step S4: Perform intelligent classification guidance on the rock wool curtain wall board according to the defect label of the rock wool curtain wall board to obtain the classification planning path of the rock wool curtain wall board; perform collaborative control of the flexible conveying system on the classification planning path of the rock wool curtain wall board to obtain the classification conveying scheme of the rock wool curtain wall board; position the rock wool curtain wall board according to the classification conveying scheme of the rock wool curtain wall board to obtain the classification storage coordinates of the rock wool curtain wall board;

[0030] Step S5: Perform detection data association on the storage coordinates of the classified rock wool curtain wall board to obtain the detection traceability link of the rock wool curtain wall board; perform digital reconstruction on the production process based on the detection traceability link to obtain the production quality management model; generate optimization suggestions for the production process of the rock wool curtain wall board according to the production quality management model to obtain the production suggestion report of the rock wool curtain wall board; apply the production suggestion report of the rock wool curtain wall board to the rock wool curtain wall board production line, and execute Steps S1 - S3 on the newly produced rock wool curtain wall board to obtain the defect label of the new rock wool curtain wall board.

[0031] In this embodiment, an industrial camera equipped with a multispectral imaging system is used to collect multispectral images of the surface of the rock wool curtain wall panel. The rock wool curtain wall panel is placed in a controlled lighting environment, and the camera is used to take pictures of the rock wool curtain wall panel from multiple angles to ensure that all details on the surface of the rock wool curtain wall panel can be captured from different angles. After shooting, the image data is transmitted to the computer system in real time and stored as a multispectral image set of the rock wool curtain wall panel. Using image processing software, such as MATLAB or the OpenCV library of Python, feature extraction is performed on the multispectral image set to obtain the initial feature map of the rock wool curtain wall panel. Frequency domain and spatial domain enhancement techniques are adopted to improve the image quality, making it easier to identify subtle defects. Through high-dimensional feature extraction and multimodal feature fusion, the features of the rock wool curtain wall panel are analyzed in depth, enhancing the defect recognition ability and improving the data processing efficiency. Geometric defect segmentation and material anomaly division are performed on the initial feature map of the rock wool curtain wall panel to obtain the geometric defect feature map and the material defect feature map of the rock wool curtain wall panel respectively. Edge detection algorithms, such as the Canny algorithm, and texture analysis techniques, such as the gray-level co-occurrence matrix (GLCM), are used to identify geometric and material defects. Feature alignment and fusion are performed on the geometric defect feature map and the material defect feature map of the rock wool curtain wall panel to obtain the comprehensive defect feature map of the rock wool curtain wall panel. Image fusion techniques, such as the Laplacian pyramid fusion method, are used to merge the geometric and material defect feature maps to provide an intuitive view for comprehensively understanding the defect state of the rock wool curtain wall panel. Defect quantification indicators are calculated for the comprehensive defect feature map of the rock wool curtain wall panel to obtain the defect feature vector of the rock wool curtain wall panel. Quantification indicators such as the area, depth, and edge complexity of the defect are calculated, and these indicators are integrated into a feature vector. Defect grading mapping is performed on the defect feature vector of the rock wool curtain wall panel to obtain the defect label of the rock wool curtain wall panel. Machine learning algorithms, such as support vector machines (SVMs), are used to grade the defects according to the preset evaluation rules. According to the defect label of the rock wool curtain wall panel, intelligent classification guidance is performed on the rock wool curtain wall panel to obtain the classification planning path of the rock wool curtain wall panel. An automated control system, such as a PLC (programmable logic controller), is used to guide the rock wool curtain wall panel to different processing areas according to the defect label. Cooperative control of the flexible conveying system is performed on the classification planning path of the rock wool curtain wall panel to obtain the classification conveying plan of the rock wool curtain wall panel. Conveying system management software is used to optimize the conveying path to ensure that the rock wool curtain wall panel can be efficiently and accurately conveyed to the designated area. According to the classification conveying plan of the rock wool curtain wall panel, positioning is performed on the rock wool curtain wall panel to obtain the classification storage coordinates of the rock wool curtain wall panel. RFID technology or a barcode scanning system is used to track the position of the rock wool curtain wall panel and record the storage coordinates. Detection data association is performed on the classification storage coordinates of the rock wool curtain wall panel to obtain the detection traceability link of the rock wool curtain wall panel. A database management system is used to integrate the storage coordinates and detection data to form a traceability link. Based on the detection traceability link, digital reconstruction of the production process is performed to obtain the production quality management model.Use data analysis software, such as the Pandas library in Python, to analyze the traceability link data and build a production quality management model. Based on the production quality management model, generate optimization suggestions for the production process of rock wool curtain wall panels and obtain a production suggestion report for rock wool curtain wall panels. Use statistical analysis and machine learning techniques, such as linear regression analysis, to identify key improvement points in the production process and generate optimization suggestions. Finally, apply the production suggestion report for rock wool curtain wall panels to the rock wool curtain wall panel production line and execute steps S1 - S3 for the newly produced rock wool curtain wall panels to obtain new rock wool curtain wall panel defect labels.

[0032] Preferably, step S1 includes the following steps:

[0033] Step S11: Collect multi - spectral images of the surface of the rock wool curtain wall panel to obtain an initial multi - spectral image set of the rock wool curtain wall panel;

[0034] Specifically, the rock wool curtain wall panel can be placed in a controlled lighting environment to ensure uniform light and avoid the influence of shadows and reflections on the imaging quality. The industrial camera used has the ability to capture images in the range from visible light to the near - infrared spectrum. Under software control, the industrial camera takes pictures of the rock wool curtain wall panel from multiple angles to ensure that all details of the surface of the rock wool curtain wall panel can be captured from different angles. For example, the camera can take pictures at four angles (0°, 90°, 180°, 270°). After each shot, the image data is transmitted to the computer system in real time and stored as an initial multi - spectral image set of the rock wool curtain wall panel. These image sets contain the response information of the rock wool curtain wall panel under different spectra.

[0035] Step S12: Perform spectral pre - calibration on the initial multi - spectral image set of the rock wool curtain wall panel to obtain a multi - spectral image set of the rock wool curtain wall panel;

[0036] Specifically, a standard reference plate with known spectral characteristics can be prepared, such as a standard white or gray plate with uniform diffuse reflection characteristics. First, place the standard reference plate under the same lighting conditions as the rock wool curtain wall panel and use the same industrial camera to take multi-angle photos of it to obtain the multi-spectral image of the reference plate. Use image processing software, such as MATLAB or the OpenCV library in Python, to analyze these images and calculate the spectral response curve of the reference plate. Compare the spectral response curve of the initial image set of the rock wool curtain wall panel with that of the reference plate to identify any spectral deviations. For example, it is found that in the spectral image at a specific wavelength, the image set of the rock wool curtain wall panel is brighter or darker than the response of the reference plate. Use this information to adjust the spectral response of the rock wool curtain wall panel image set to match that of the reference plate. Adopt a linear or non-linear calibration model to perform pixel-by-pixel adjustment of the rock wool curtain wall panel image set according to the spectral response of the reference plate. For example, if it is found that the response of the rock wool curtain wall panel image set at wavelength λ1 is generally 10% higher than that of the reference plate, then multiply the pixel values at all wavelengths λ1 by 0.9 for correction. Finally, obtain the calibrated multi-spectral image set of the rock wool curtain wall panel.

[0037] Step S13: Perform frequency domain enhancement on the multi-spectral image set of the rock wool curtain wall panel to obtain a frequency domain enhanced multi-spectral image set;

[0038] Specifically, the Fourier Transform can be used to transform the images in each spectral channel of the multi-spectral image set of the rock wool curtain wall panel, converting the image in the spatial domain to the frequency domain. In the frequency domain, analyze and adjust the frequency components of the image. For example, enhance the high-frequency components in the image through a High-pass Filter, which helps to highlight the edge and detail information in the image. In specific operations, design a mask for the high-pass filter, whose size matches the frequency resolution of the image, with the center set to zero and the periphery as positive values. Apply the filter to the frequency domain images of each spectral channel, and obtain the enhanced frequency domain image by adding the filtered frequency domain image to the original frequency domain image. Finally, perform the Inverse Fourier Transform on the enhanced frequency domain images to convert them back to the spatial domain, obtaining the frequency domain enhanced multi-spectral image set. This process can be implemented using image processing software such as MATLAB or the OpenCV library in Python.

[0039] Step S14: Perform spatial domain enhancement on the frequency domain enhanced multi-spectral image set to obtain a frequency domain - spatial domain enhanced multi-spectral image set;

[0040] Specifically, histograms can be calculated for the images of each spectral channel after frequency-domain enhancement. The histogram shows the number of pixels for each gray value in the image. The Cumulative Distribution Function (CDF) is calculated based on the histogram, and an equalization function is applied to adjust the intensity values of the original image, making the histogram distribution of the output image more uniform, thereby enhancing the global contrast of the image. These operations can be implemented through image processing software or image processing libraries in programming languages, such as MATLAB, OpenCV, or the PIL library. Through these spatial-domain enhancement operations, a frequency-domain - spatial-domain enhanced multi-spectral image set can be obtained.

[0041] Step S15: Perform high-dimensional feature extraction on the frequency-domain - spatial-domain enhanced multi-spectral image set to obtain a high-dimensional feature set of the rock wool curtain wall board, and perform multi-modal feature fusion on the high-dimensional feature set of the rock wool curtain wall board to obtain an initial feature map of the rock wool curtain wall board.

[0042] Specifically, a machine learning library, such as scikit-learn in Python, can be used for high-dimensional feature extraction. For high-dimensional feature extraction, the Principal Component Analysis (PCA) method is adopted. The specific operations are as follows: First, each pixel point in the multi-spectral image set is regarded as a point in a high-dimensional space, where the value of each spectral channel corresponds to one dimension. The PCA algorithm is applied to these high-dimensional data to extract the main components, which can explain the largest variance in the data. By selecting the first few main components, a feature set with reduced dimensions can be obtained, and the data in these feature sets can represent the most critical information in the original data set. Next, multi-modal feature fusion is performed. In this step, different types of features need to be combined, such as texture features, shape features, and spectral features. Deep learning methods, such as Convolutional Neural Networks (CNN), are used to automatically learn the complex relationships between the features. A CNN model is constructed. The input layer receives the multi-spectral image set, and the intermediate layer automatically extracts features through multiple convolutional layers and pooling layers. Finally, the fused feature map is output through the fully connected layer. During the training process, the CNN model adjusts the weights through the backpropagation algorithm to minimize the prediction error. After training is completed, the enhanced multi-spectral image set is propagated forward using the model to obtain the high-dimensional feature set of the rock wool curtain wall board, and these features are fused to finally obtain the initial feature map of the rock wool curtain wall board. This feature map contains the key information extracted from the multi-spectral image set.

[0043] The present invention enhances the comprehensive recognition ability of the surface features of rock wool curtain wall panels through multi-spectral image acquisition, ensures the accuracy of the image set, and improves the image quality through frequency domain and spatial domain enhancement techniques, making it easier to identify subtle defects. By extracting high-dimensional features and fusing multi-modal features, the features of rock wool curtain wall panels are deeply analyzed, the defect recognition ability is enhanced, and the data processing efficiency is improved.

[0044] Preferably, step S2 includes the following steps:

[0045] Step S21: Extract geometric shape features from the initial feature map of the rock wool curtain wall panel to obtain a geometric feature descriptor of the rock wool curtain wall panel;

[0046] Specifically, an edge detection algorithm such as the Canny edge detector can be applied to the initial feature map of the rock wool curtain wall panel. The Canny algorithm smooths the image through a Gaussian filter, then uses Sobel operators with two different thresholds to determine the gradient magnitude and direction of the image, and finally determines strong edges through non-maximum suppression and double-threshold detection. Through the Canny algorithm, the edge information in the feature map of the rock wool curtain wall panel can be obtained, and these edge information form the basis of the geometric shape features of the rock wool curtain wall panel. Next, the Hough Transform is used to detect geometric shapes such as straight lines and circles in the feature map. The Hough Transform is a feature extraction technique that can map lines or circles in the image space to the parameter space, thereby simplifying the recognition of shapes. For example, for straight line detection, each edge point in the image space is mapped to a series of straight lines in the parameter space; for circle detection, the edge points are mapped to circles in the parameter space. By the peak in the accumulator space, geometric features such as straight lines and circles in the image can be identified. Through these operations, a geometric feature descriptor of the rock wool curtain wall panel can be obtained, including edge information, straight lines, circles, etc.

[0047] Step S22: Detect the defect edges of the geometric feature descriptor of the rock wool curtain wall panel to obtain the geometric defect contour of the rock wool curtain wall panel, and identify the defect connectivity of the geometric defect contour of the rock wool curtain wall panel to obtain defect network connectivity data;

[0048] Specifically, the geometric feature descriptors of the rock wool curtain wall panels can be binarized. After binarization, morphological operations such as dilation and erosion are used to enhance the continuity of the defect edges in the image. The dilation operation can increase the area of the white region (i.e., the defect edges) in the image, while the erosion operation can reduce the area of the white region. The combination of these two operations can strengthen the defect edges and fill small gaps. The contour detection algorithm, such as the findContours function in OpenCV, is used to detect the contours in the processed binary image. This function can find the boundaries of all the contours in the image and return their coordinates. By analyzing these contours, the geometric defect contours of the rock wool curtain wall panels can be obtained. Finally, connectivity analysis is performed on the detected defect contours. The connected component algorithm in graph theory is used to identify which defect edges are connected to each other. By constructing a graph where each contour point is a node and the connectivity between adjacent points is an edge, the size and shape of each connected component are calculated. This provides defect network connectivity data, which indicates the spatial relationship between the defects.

[0049] Step S23: Identify the defect morphological dimensions of the geometric defect contours of the rock wool curtain wall panels to obtain defect geometric dimension distribution data, and calculate the defect spatial entropy density of the geometric defect contours of the rock wool curtain wall panels to obtain defect spatial distribution complexity data;

[0050] Specifically, image processing and analysis software such as MATLAB or the SciPy library in Python can be used to identify the defect morphological dimensions. First, it is achieved by calculating the geometric properties of the geometric defect contours of the rock wool curtain wall panels. For example, basic geometric parameters such as the length, area, perimeter, and convex hull of the contour are used. For example, the approxPolyDP function in the OpenCV library is used to approximate the shape of the contour and calculate its perimeter and area. For more complex shapes, the Hu moments of the contour are calculated. These are moments that describe shape invariant properties and can be used to distinguish different shapes. Next, the calculation of the defect spatial entropy density is carried out. The spatial entropy density is an indicator describing the complexity of the defect distribution and is calculated using the Local Binary Pattern (LBP) of the image. LBP is a texture descriptor that can be used to capture local texture information in the image. The LBP operator is applied to the geometric defect contours of the rock wool curtain wall panels to generate an LBP image, and then the distribution of the LBP values of each pixel is calculated to obtain the defect spatial distribution complexity data. Through these operations, the defect geometric dimension distribution data and the defect spatial distribution complexity data of the rock wool curtain wall panels can be obtained.

[0051] Step S24: Perform feature clustering on the geometric defect profiles of the rock wool curtain wall panels based on the defect network connectivity data, defect geometric dimension distribution data, and defect spatial distribution complexity data to obtain the geometric defect pattern classification data of the rock wool curtain wall panels, and perform geometric defect characteristic annotation on the initial feature map of the rock wool curtain wall panels according to the geometric defect pattern classification data of the rock wool curtain wall panels to obtain the geometric defect feature map of the rock wool curtain wall panels;

[0052] Specifically, a machine learning library such as scikit-learn in Python can be used for feature clustering. First, integrate the defect network connectivity data, defect geometric dimension distribution data, and defect spatial distribution complexity data into a feature vector. Each defect profile corresponds to a feature vector that contains all relevant features. Select a clustering algorithm such as K-means or DBSCAN to cluster these feature vectors. For example, use the K-means algorithm, which is a centroid-based clustering method that iteratively optimizes the cluster centroids to minimize the sum of the squared distances within the clusters. Determine the optimal number of clusters K value according to the elbow method or silhouette coefficient, and then run the K-means algorithm to cluster the feature vectors to obtain the geometric defect pattern classification data of the rock wool curtain wall panels. After clustering, perform geometric defect characteristic annotation on the initial feature map of the rock wool curtain wall panels. Use the drawing functions in the OpenCV library to mark the regions of each cluster on the initial feature map. For example, use different colors or labels to represent different defect patterns, and finally obtain the geometric defect feature map of the rock wool curtain wall panels, where each defect region is annotated with the corresponding defect pattern.

[0053] Step S25: Perform material anomaly division on the initial feature map of the rock wool curtain wall panels to obtain the material defect feature map of the rock wool curtain wall panels;

[0054] Specifically, for the detailed implementation process of this embodiment, please refer to the sub-steps of Step S25.

[0055] Step S26: Perform weighted fusion of the defect features of the geometric defect feature map of the rock wool curtain wall panels and the material defect feature map of the rock wool curtain wall panels to obtain the comprehensive defect feature map of the rock wool curtain wall panels.

[0056] Specifically, the weights of the geometric defect feature map and the material defect feature map of the rock wool curtain wall board can be determined. These weights can be set according to the severity of the defects, the importance of the defect types, or prior knowledge. For example, if geometric defects are considered more important than material defects, a larger weight, such as 0.6, can be assigned to the geometric defect feature map, and a smaller weight, such as 0.4, can be assigned to the material defect feature map. First, multiply each pixel value of the geometric defect feature map by its corresponding weight of 0.6. Similarly, multiply each pixel value of the material defect feature map by its corresponding weight of 0.4. Then, add the corresponding pixel values of the two weighted feature maps to obtain the pixel values of the comprehensive feature map. In this way, the comprehensive value of each pixel point is obtained by the weighted fusion of the feature values of geometric defects and material defects. For example, assume that in a certain area, the pixel value of the geometric defect feature map is 150, and the pixel value of the material defect feature map is 100. According to the weights, multiply the pixel value of the geometric defect feature map by 0.6 to get 90, and multiply the pixel value of the material defect feature map by 0.4 to get 40. Add these two values to obtain the pixel value of this area in the comprehensive feature map as 130. In this way, calculate each pixel point of the entire feature map, and finally obtain the comprehensive feature map of the defects of the rock wool curtain wall board.

[0057] Through geometric defect recognition and defect edge detection, the present invention can provide accurate basic data for the geometric defects of the rock wool curtain wall board. Furthermore, through the evaluation of defect network connectivity, the mutual relationship between defects can be understood. The understanding of the complexity of the geometric dimension distribution and spatial distribution of defects is enhanced through multi-dimensional defect analysis, while feature clustering and pattern classification provide a basis for the effective classification of different defects, guiding subsequent defect processing. Through the generation of the comprehensive defect feature map, through the weighted fusion of the geometric and material defect feature maps, an intuitive view of the defect state of the rock wool curtain wall board is provided.

[0058] Preferably, step S25 includes the following steps:

[0059] Step S251: Extract the material texture features of the initial feature map of the rock wool curtain wall board to obtain the material feature mapping data of the rock wool curtain wall board;

[0060] Specifically, the Gray Level Co-occurrence Matrix (GLCM) method can be used to extract material texture features. The initial feature map of the rock wool curtain wall panel is converted into a grayscale image because GLCM is calculated based on grayscale images. Select a specific direction and distance parameter, which define the relative positions of pixel pairs in GLCM. For example, select a direction of 0° and a distance of 1 pixel, which will calculate the grayscale relationship of adjacent pixels in the horizontal direction. Next, calculate the GLCM matrix, which records the occurrence frequencies of all grayscale pairs at the given direction and distance. From the GLCM matrix, various texture features can be extracted, such as contrast, uniformity, energy, and entropy. These features can describe the texture characteristics of the rock wool curtain wall panel material, such as the roughness and uniformity of the texture. Finally, map these texture features back to the initial feature map of the rock wool curtain wall panel, create a corresponding material feature mapping data for each pixel position, and finally obtain the rock wool curtain wall panel material feature mapping data.

[0061] Step S252: Identify material anomalies in the rock wool curtain wall panel material feature mapping data to obtain the rock wool curtain wall panel material anomaly mapping data, and calculate the material feature gradient for the initial feature map of the rock wool curtain wall panel based on the rock wool curtain wall panel material anomaly mapping data to obtain the rock wool curtain wall panel material feature gradient;

[0062] Specifically, machine learning algorithms can be used to identify material anomalies. The rock wool curtain wall panel material feature mapping data includes texture features such as contrast and uniformity. Use the Support Vector Machine (SVM) algorithm, which is a commonly used classification algorithm, to identify material anomalies in the rock wool curtain wall panel. First, a training data set is required, which contains the material feature mapping data of normal and abnormal rock wool curtain wall panels. Use this training data set to train an SVM classifier. After training, apply the SVM classifier to the material feature mapping data of the rock wool curtain wall panel to identify the abnormal areas. For the identified abnormal areas, calculate the material feature gradient. This can be achieved through the Sobel operator, which is an algorithm for edge detection and can calculate the gradient magnitude and direction of each pixel in the image. Apply the Sobel operator to the abnormal areas to obtain the material feature gradient, which reflects the degree of texture change in the abnormal areas. Finally, integrate this gradient information into the initial feature map of the rock wool curtain wall panel, create a corresponding material feature gradient value for each abnormal pixel position, and finally obtain the rock wool curtain wall panel material feature gradient.

[0063] Step S253: Quantify the anomaly intensity of the rock wool curtain wall panel material feature gradient to obtain the rock wool curtain wall panel material defect intensity data;

[0064] Specifically, image processing software such as MATLAB or the OpenCV library of Python can be used to process the gradient data of the material characteristics of the rock wool curtain wall panel. Specifically, quantitative analysis is performed on the gradient image obtained through the Sobel operator. The value of each pixel point in the gradient image represents the gradient magnitude at that point, that is, the severity of the texture change. First, a threshold is determined, and this threshold can be set based on historical data, expert experience, or statistical analysis. For example, it is found that in the normal material of the rock wool curtain wall panel, the gradient value is usually lower than a certain specific value, while the gradient value in the abnormal area is significantly higher than this value. Set the threshold to this specific value, and then compare each pixel value in the gradient image with the threshold. For pixels exceeding the threshold, it is considered that they represent areas with a higher abnormal intensity. By calculating the difference between each pixel value and the threshold and normalizing it to a specific range (such as 0 to 1), the intensity of these abnormalities is quantified, and each pixel point has a quantization value corresponding to the abnormal intensity. Record these quantization values to form the material defect intensity data of the rock wool curtain wall panel. These data provide a measure of the abnormal intensity for each pixel point.

[0065] Step S254: Perform material defect grading mapping on the material defect intensity data of the rock wool curtain wall panel to obtain the material defect labels of the rock wool curtain wall panel;

[0066] Specifically, statistical analysis can be performed on the material defect intensity data to determine the distribution of different intensity levels. For example, it is found that the defect intensity data roughly follows a normal distribution, or there are several obvious peaks, and these peaks represent different types of material defects. Based on these statistical information, several different defect intensity levels are defined, such as: "slight", "medium", and "severe". Each level corresponds to a specific intensity value range. For example, "slight" corresponds to a range close to the normal gradient value, "medium" corresponds to a range moderately deviated from the normal value, and "severe" corresponds to a range significantly higher than the normal value. Use these level definitions to map the defect intensity data of each pixel point. For each pixel point, classify it into the corresponding level according to its defect intensity value and assign a corresponding defect label. For example, if the defect intensity value of a pixel point is 0.8, classify it into the "severe" level and assign a "severe" label. Finally, generate a material defect label map of the rock wool curtain wall panel, where each pixel point is marked with its corresponding defect level, and finally obtain the material defect labels of the rock wool curtain wall panel. These labels intuitively represent the severity of the defects in each part of the rock wool curtain wall panel.

[0067] Step S255: Perform material property annotation on the initial feature map of the rock wool curtain wall panel according to the material defect labels of the rock wool curtain wall panel to obtain the material defect feature map of the rock wool curtain wall panel.

[0068] Specifically, image processing software such as Adobe Photoshop or professional image analysis software such as MATLAB can be used to process the initial feature map of the rock wool curtain wall panel and the corresponding material defect label data. Assume that these label data already exist in numerical or categorical form and correspond to each pixel point in the initial feature map. The rock wool curtain wall panel will be color-coded according to the severity of the material defects. For example, blue can be assigned to "minor" defects, yellow to "moderate" defects, and red to "severe" defects. Such color coding helps to quickly visually distinguish defects of different degrees. First, load the initial feature map of the rock wool curtain wall panel into the image processing software. According to the material defect label corresponding to each pixel point, use the drawing tools or programming scripts of the software to add corresponding color markings on the initial feature map. For example, if an area is marked as a "severe" defect, cover that area with red; if an area is marked as a "minor" defect, cover that area with blue. After completing the marking of all pixel points, a material defect feature map of the rock wool curtain wall panel can be obtained. This image visually shows the material defect conditions in each area of the rock wool curtain wall panel, where different colors represent defects of different degrees.

[0069] The present invention significantly enhances the recognition ability of material defects by extracting the material texture features of the rock wool curtain wall panel, and improves the detail level of defect detection by calculating the material feature gradient. The quantification of abnormal intensity provides a quantification method for evaluating the severity of material defects, while the material defect grading mapping realizes more refined defect management. The generation of material defect labels and the optimization of material property annotation optimize defect annotation and feature map generation, improving the pertinence and effectiveness of defect processing.

[0070] Preferably, step S3 includes the following steps:

[0071] Step S31: Perform pixel-level defect area statistics on the comprehensive defect feature map of the rock wool curtain wall panel to obtain the defect coverage area of the rock wool curtain wall panel;

[0072] Specifically, image processing software such as MATLAB or the OpenCV library of Python can be used to process the comprehensive feature map of the defects of the rock wool curtain wall board. The comprehensive feature map of the defects is loaded into the software and converted into a binary image. In the binary image, the defect area is marked as white (or any specified non-zero value), and the non-defect area is marked as black (or zero value). This step can be achieved by the threshold segmentation method, that is, setting a threshold, regarding the pixel points higher than the threshold as the defect area, and the pixel points lower than the threshold as the non-defect area. Pixel counting is performed on the binary image to statistically calculate the total area of the defect area. In MATLAB, the bwprop function is used to calculate the properties of the white pixels in the binary image, such as area, perimeter, etc. For example, if it is found that there are 10,000 white pixels in the binary image and each pixel represents an area of 1 square millimeter, then the defect coverage area of the rock wool curtain wall board is 10,000 square millimeters. These statistical data are recorded to obtain the accurate value of the defect coverage area of the rock wool curtain wall board.

[0073] Step S32: Extract the morphological defect depth features from the comprehensive feature map of the defects of the rock wool curtain wall board to obtain the defect depth distribution data of the rock wool curtain wall board;

[0074] Specifically, professional image analysis software such as MATLAB or ImageJ can be used to process the comprehensive feature map of the defects of the rock wool curtain wall board to extract the defect depth features. First, edge detection is performed on the comprehensive feature map of the defects to determine the boundaries of the defect areas. In MATLAB, the edge function is used to apply the Canny edge detection algorithm, and defect depth analysis is performed within the determined defect boundaries. The morphological gradient method is used to estimate the depth of the defects. Specifically, a morphological gradient operator such as the Sobel operator or the Prewitt operator is applied to the defect area to calculate the gradient magnitude of the defect edges, which reflects the change in the defect depth. Statistical analysis is performed on the gradient magnitudes of the entire defect area. For example, calculate the average value, median value, maximum value, and minimum value of the gradient magnitudes, as well as the histogram of the gradient magnitudes. These statistical data provide comprehensive information about the defect depth distribution. Through these operations, the defect depth distribution data of the rock wool curtain wall board are finally obtained.

[0075] Step S33: Perform scale normalization on the defect depth distribution data of the rock wool curtain wall board to obtain the defect depth index set of the rock wool curtain wall board;

[0076] Specifically, the min-max normalization method can be adopted for scale normalization, and the data can be scaled to a specified interval, such as [0, 1]. Determine the maximum depth value and the minimum depth value in the defect depth distribution data of the rock wool curtain wall board. Then, for each depth value in the dataset, the following formula is used for normalization processing: Normalized depth value ; In this way, each original depth value is converted into a normalized value between 0 and 1, where 0 represents the minimum depth and 1 represents the maximum depth. In this manner, a set of normalized depth metrics can be obtained, and each value in this set represents the relative position of the corresponding defect depth in the overall depth distribution.

[0077] Step S34: Evaluate the topological complexity of the edge features of the comprehensive defect map of the rock wool curtain wall panel to obtain a dataset for evaluating the defect edge complexity;

[0078] Specifically, image analysis software such as MATLAB or ImageJ can be used to process the comprehensive defect map of the rock wool curtain wall panel to evaluate the topological complexity of the defect edges. First, perform edge detection on the comprehensive defect map to determine the boundaries of the defect areas. Use the Canny edge detection algorithm to analyze the detected edges to evaluate their complexity. The curvature of the edges and the number of bifurcation points can be used as indicators to measure the edge complexity. The curvature can be evaluated by calculating the second derivative of the edge points, and the bifurcation points can be identified by detecting changes in the edge width. First, calculate the second derivative of the edge image to obtain the curvature information of the edges. Count the points with significant width changes in the edge image, and these points are the bifurcation points. Record the number of these curvatures and bifurcation points, and calculate their statistical data such as the average curvature and the bifurcation point density. Finally, organize these statistical data into a dataset for evaluating the defect edge complexity.

[0079] Step S35: Calculate the defect quantification metrics for the comprehensive defect map of the rock wool curtain wall panel based on the defect coverage area of the rock wool curtain wall panel, the depth metric set of the rock wool curtain wall panel defects, and the dataset for evaluating the defect edge complexity to obtain the defect feature vector of the rock wool curtain wall panel;

[0080] Specifically, for the detailed implementation process of this embodiment, please refer to the sub-steps of Step S35.

[0081] Step S36: Perform defect classification mapping on the defect feature vector of the rock wool curtain wall panel according to the preset evaluation rules to obtain the defect label of the rock wool curtain wall panel.

[0082] Specifically, a set of preset evaluation rules can be defined, which are based on the various components of the defect feature vector, including the defect coverage area, the defect depth topology variation coefficient, and the defect edge topology variation coefficient. For example, the following rules are set: If the defect coverage area is less than 100 square millimeters and both the defect depth and the edge topology variation coefficients are below 0.1, the defect is classified as "minor"; if the defect coverage area is between 100 and 500 square millimeters and any one of the topology variation coefficients is between 0.1 and 0.3, it is "moderate"; if the defect coverage area exceeds 500 square millimeters or any one of the topology variation coefficients exceeds 0.3, it is "severe". A decision support system or a simple logical judgment algorithm is used to process the defect feature vector of the rock wool curtain wall panel. Suppose there is a defect feature vector with a defect coverage area of 200 square millimeters, a defect depth topology variation coefficient of 0.25, and a defect edge topology variation coefficient of 0.2. According to the preset evaluation rules, the coverage area of this defect falls within the range of 100 to 500 square millimeters, and the defect depth topology variation coefficient exceeds 0.1 but is below 0.3, so it is classified as "moderate". Such evaluations and classifications are performed on all defect feature vectors, and finally the label of each defect is obtained. For example, a table is created that lists the number of each defect, the corresponding feature vector values, and the defect grade obtained according to the preset rules. This table will serve as a record of the defect labels of the rock wool curtain wall panel.

[0083] The present invention realizes the precise quantification of the defects of the rock wool curtain wall panel through pixel-level defect area statistics and morphological defect depth feature extraction. By performing scale normalization processing, the comparison benchmark for the defect depth data of different rock wool curtain wall panels is unified, enhancing the standardization and consistency of the evaluation. Through the evaluation of the topological complexity of the edge features, the understanding of the defect shape and edge complexity is deepened, and the comprehensive defect feature vector integrates the coverage area, depth, and edge complexity, laying a foundation for precise classification and evaluation. The efficiency and effectiveness of defect management are improved through the defect classification under the preset evaluation rules, and at the same time, the severity of the defects is accurately evaluated.

[0084] Preferably, step S35 includes the following steps:

[0085] Step S351: Calculate the standard deviation of the defect depth index set and the defect edge complexity evaluation data set of the rock wool curtain wall panel respectively to obtain the defect depth standard deviation and the defect edge complexity standard deviation;

[0086] Specifically, the defect depth index set of the rock wool curtain wall board and the defect edge complexity evaluation data set can be input into the software. In Excel, the STDEV function is used to calculate the standard deviation. For example, if there is a data set containing 100 defect depth values, select all the values and then apply the STDEV function to obtain the standard deviation of the entire data set. This value represents the degree of dispersion of the defect depth value distribution. For the defect edge complexity evaluation data set, repeat the same steps. Suppose there is a data set containing 80 edge complexity values, use the STDEV function to calculate the standard deviation of these values, obtain the standard deviation of the defect edge complexity, and finally obtain the standard deviation of the defect depth and the standard deviation of the defect edge complexity, which describe the degree of variation of the defect depth and edge complexity of the rock wool curtain wall board.

[0087] Step S352: Calculate the mean values of the defect depth index set of the rock wool curtain wall board and the defect edge complexity evaluation data set respectively to obtain the average defect depth and the average defect edge complexity;

[0088] Specifically, in Excel, the AVERAGE function can be used to calculate the mean value. For example, for the defect depth index set, select all the depth values and then apply the AVERAGE function or the mean function to obtain the average value of these values. This value represents the average level of the defect depth of the rock wool curtain wall board. For the defect edge complexity evaluation data set, perform the same operation. Suppose there is a data set containing 80 edge complexity values, select all the values and then apply the AVERAGE function or the mean function to calculate the average value of these values, obtain the average defect edge complexity, and finally obtain the average defect depth and the average defect edge complexity.

[0089] Step S353: Calculate the coefficient of variation of the defect depth index set of the rock wool curtain wall board based on the standard deviation of the defect depth and the average defect depth to obtain the topological coefficient of variation of the defect depth;

[0090] Specifically, Excel or the NumPy library in Python can be used to perform the following operations: First, determine the specific values of the standard deviation of the defect depth and the average defect depth. Suppose the standard deviation of the defect depth is 0.8 mm and the average defect depth is 4.0 mm. Apply the formula for the coefficient of variation: the topological coefficient of variation of the defect depth ; This value of 0.2 represents the relative degree of dispersion of the defect depth, that is, the standard deviation is 20% of the mean. Record this calculation result as the topological coefficient of variation of the defect depth, which reflects the consistency and stability of the defect depth of the rock wool curtain wall board.

[0091] Step S354: Calculate the coefficient of variation for the defect depth index set of the rock wool curtain wall panel based on the standard deviation and average value of the defect edge complexity, obtain the defect edge topology coefficient of variation, and record the defect coverage area, defect depth topology coefficient of variation, and defect edge topology coefficient of variation of the rock wool curtain wall panel as the defect feature vector of the rock wool curtain wall panel.

[0092] Specifically, it can be assumed that the standard deviation of the defect edge complexity is 0.5 and the average value of the defect edge complexity is 2.5. Use the same method as in step S353 to calculate the coefficient of variation: the defect edge topology coefficient of variation ; this value of 0.2 represents the relative dispersion degree of the defect edge complexity, that is, the standard deviation is 20% of the mean. Record this calculation result as the defect edge topology coefficient of variation. Finally, integrate the defect coverage area, defect depth topology coefficient of variation, and defect edge topology coefficient of variation of the rock wool curtain wall panel together to form a defect feature vector of the rock wool curtain wall panel. For example, if the defect coverage area is 300 square millimeters, the defect depth topology coefficient of variation is 0.2, and the defect edge topology coefficient of variation is also 0.2, then the defect feature vector of the rock wool curtain wall panel can be expressed as (300, 0.2, 0.2).

[0093] The present invention enhances the analysis ability of the statistical characteristics of defects by calculating the standard deviation and mean value of the defect depth and edge complexity of the rock wool curtain wall panel, and can understand the distribution characteristics of defect data in detail. The variability of defects is quantified through the calculation of the coefficient of variation, and the consistency and stability of defects are evaluated. By introducing the defect depth topology coefficient of variation and the defect edge topology coefficient of variation, the description accuracy of the defect feature vector is improved, making the defect features more detailed and comprehensive. Combining these indicators can more accurately identify and distinguish different degrees of defects, improve the accuracy of defect identification, provide an objective and scientific basis for defect grading, and enhance the scientificity and accuracy of defect grading.

[0094] Preferably, step S4 includes the following steps:

[0095] Step S41: Collect the distribution map of the transportation route of the rock wool curtain wall panel to obtain the transportation route network topology diagram;

[0096] Specifically, Geographic Information System (GIS) software or automated logistics planning software can be used to collect the distribution map of transportation routes. These software can obtain detailed information about transportation routes from on-site measurements or existing factory layout plans, including the location, length, connection points, etc. of the routes. For example, using the mapping function of GIS software, through aerial photography by drones or ground measurements, a detailed route map between the rock wool curtain wall panel production line and the storage area is obtained. Inputting this route information into the logistics planning software, the software will generate a network topology map based on the connection relationships of the routes. In this topology map, each node represents a connection point of a transportation route, such as a loading and unloading area, a turning point, etc., and each edge represents a route segment between two nodes. For example, create a node in the software to represent the exit of the production line and another node to represent the entrance of the storage area, and then connect these nodes according to the actual transportation path to form a transportation route network topology map.

[0097] Step S42: Perform data preprocessing on the transportation route network topology map to obtain a transportation path feature vector, where the data preprocessing includes path node weight calculation, path connection feature extraction, and transportation path capacity evaluation;

[0098] Specifically, logistics analysis software or data processing software, such as the NetworkX library in Python, can be used to perform data preprocessing on the transportation route network topology map. First, calculate the weights of path nodes, which can be based on factors such as the transportation flow, loading and unloading efficiency, or delay probability of the nodes. For example, set the weight of a node as the number of rock wool curtain wall panels passing through the node per day multiplied by the average loading and unloading time of the node. Extract path connection features, which include the length, slope, turning radius, etc. of the routes. For example, use the CAD layout map of the factory to obtain the exact dimensions of these routes and calculate the length and slope of each route. Then, evaluate the capacity of the transportation paths, which involves the maximum bearing capacity of each route, such as the maximum load capacity or the maximum throughput. For example, determine the capacity of each route based on the specifications of the transportation vehicles and the physical conditions of the routes. Finally, integrate these data into a transportation path feature vector. For example, for a specific transportation path, a feature vector can be obtained, including information such as node weight, route length, slope, and capacity.

[0099] Step S43: Construct the hierarchical transportation constraint conditions for the rock wool curtain wall panels according to the defect labels of the rock wool curtain wall panels to obtain the hierarchical transportation path selection criteria;

[0100] Specifically, different transportation priorities and processing requirements can be defined according to the defect labels of the rock wool curtain wall panels, such as "slight", "medium", and "severe". For example, the rock wool curtain wall panels with "severe" defects require the shortest transportation time for quick repair or scrapping, while the rock wool curtain wall panels with "slight" defects can follow the conventional transportation route. Use logistics planning software to define these constraints. In the software, set specific transportation parameters for each defect level, such as the maximum transportation time, priority level, and allowed transportation routes. For example, set that the rock wool curtain wall panels with "severe" defects must reach the processing area within 10 minutes, while the maximum transportation time for the rock wool curtain wall panels with "slight" defects can be relaxed to 30 minutes. Construct hierarchical transportation constraints based on these parameters. For example, for the rock wool curtain wall panels with "severe" defects, set constraints in the software to only allow the selection of transportation routes that can reach within 10 minutes. For the rock wool curtain wall panels with "medium" and "slight" defects, set different time limits and route selection conditions according to their priorities. Finally, obtain the hierarchical transportation route selection criteria based on these constraints.

[0101] Step S44: Optimize the calculation of the target transportation route for the rock wool curtain wall panels based on the hierarchical transportation route selection criteria and the transportation route feature vector to obtain the classified planning routes for the rock wool curtain wall panels;

[0102] Specifically, path optimization algorithms such as the Dijkstra algorithm can be used to calculate the best transportation route for each rock wool curtain wall panel. These algorithms can find the shortest path from the starting point to the ending point considering factors such as path length, time, and other constraints. First, input the hierarchical transportation route selection criteria and the transportation route feature vector into the path optimization software. For example, there is a rock wool curtain wall panel with "severe" defects that needs to be transported from the production line to the processing area. Set the parameters of the algorithm according to its constraints, such as the maximum transportation time of 10 minutes. Then, run the path optimization algorithm, and the algorithm will calculate the best transportation route based on the input criteria and feature vector. For example, the algorithm will consider factors such as the length of the path, traffic conditions, and the capacity of the path, and finally determine a path that not only meets the time requirements but also maximizes the transportation efficiency. Finally, obtain the classified planning routes for the rock wool curtain wall panels, and these routes provide the optimal transportation solutions for the rock wool curtain wall panels of each defect level. For example, plan the shortest and fastest path for the rock wool curtain wall panels with "severe" defects, and plan the most cost-effective path for the rock wool curtain wall panels with "slight" defects. Ultimately, a classified planning route for the rock wool curtain wall panels that comprehensively considers the defect level and transportation efficiency can be obtained.

[0103] Step S45: Perform collaborative control of the flexible transportation system on the classified planning routes for the rock wool curtain wall panels through the end effector of the flexible transportation system to obtain the classified transportation plan for the rock wool curtain wall panels;

[0104] Specifically, for the detailed implementation process of this embodiment, please refer to the sub-steps of step S45.

[0105] Step S46: Automatically transport the rock wool curtain wall panels according to the classified transportation plan for rock wool curtain wall panels, and position the rock wool curtain wall panels after the operation is completed to obtain the classified storage coordinates of the rock wool curtain wall panels.

[0106] Specifically, it is possible to

[0107] The present invention optimizes the transportation path of the rock wool curtain wall panels by collecting the transportation route distribution map and data preprocessing, significantly improving the logistics efficiency, reducing the transportation time and cost. By utilizing the transportation path feature vector and the hierarchical transportation path selection criterion, the path planning becomes more scientific and reasonable, improving the quality of the transportation plan. By constructing the hierarchical transportation constraint conditions according to the defect labels of the rock wool curtain wall panels, the refined logistics management of the rock wool curtain wall panels is realized, ensuring the reasonable distribution and processing of different grade products. By applying the flexible transportation system and the automatic transportation operation, the automation level of the system is improved, reducing the manual intervention, lowering the operation error rate, optimizing the storage layout at the same time, and improving the utilization rate and access efficiency of the storage space. By quickly and accurately classifying and planning the path and transportation plan, the flexibility and response speed of production are improved, the automatic transportation reduces the operation cost, enhances the operation safety, and reduces the work injury accidents.

[0108] Preferably, step S45 includes the following steps:

[0109] Step S451: Perform real-time three-dimensional coordinate tracking on the rock wool curtain wall panels to obtain the spatial dynamic coordinates of the rock wool curtain wall panels.

[0110] Specifically, a three-dimensional coordinate tracking system integrating multiple sensors can be used. For example, a system that combines a laser scanner, a camera, and an infrared sensor can be used to track the position and attitude of the rock wool curtain wall panel. These sensors can capture the three-dimensional coordinates of the rock wool curtain wall panel in real time. The sensors are installed at key positions on the production line to ensure that they can cover the entire transportation path of the rock wool curtain wall panel. When the rock wool curtain wall panel enters the working range of the sensors, the system starts to collect data. For example, the laser scanner can measure the distance between the rock wool curtain wall panel and the scanner, the camera can capture the image of the rock wool curtain wall panel and identify its feature points, and the infrared sensor can detect the temperature change of the rock wool curtain wall panel. Data processing software, such as MATLAB or the SciPy library of Python, is used to integrate the data of these sensors. The software processes the sensor signals through algorithms and calculates the real-time three-dimensional coordinates of the rock wool curtain wall panel. For example, the software can use the feature points captured by the camera and the known camera parameters to calculate the position and attitude of the rock wool curtain wall panel in space through computer vision algorithms (such as solving the PnP problem). Finally, these coordinate data are updated in real time to the information system of the factory so that other systems and operators can access the spatial dynamic coordinates of the rock wool curtain wall panel in real time.

[0111] Step S452: Plan the grasping path of the rock wool curtain wall panel according to the spatial dynamic coordinates of the rock wool curtain wall panel to obtain the grasping path of the rock wool curtain wall panel;

[0112] Specifically, path planning software, such as Autodesk Robotics Structural Analysis, or professional path planning algorithms, such as the RRT (Rapidly-Exploring Random Trees) algorithm, can be used to plan the grasping path of the robotic arm. The real-time three-dimensional coordinates of the rock wool curtain wall panel are input into the path planning software. The software calculates the optimal path from the current position of the robotic arm to the rock wool curtain wall panel based on these coordinate data and the three-dimensional model of the factory layout. For example, the software can consider the obstacles in the factory, the operating status of other equipment, and the size of the rock wool curtain wall panel to plan a path that avoids collisions and has the shortest distance. Then, the software generates a detailed grasping path, including the coordinates of each node on the path and the corresponding attitude of the robotic arm. For example, the path includes a series of steps of moving from the starting position of the robotic arm to a point directly above the rock wool curtain wall panel, then descending to a certain height above the rock wool curtain wall panel, and finally performing the grasping action. Finally, this grasping path is transmitted to the robotic arm control system.

[0113] Step S453: Regulate the conveyor belt speed of the flexible conveying system according to the classification planning path of the rock wool curtain wall panel and the defect label of the rock wool curtain wall panel to obtain the conveying rate parameter of the rock wool curtain wall panel;

[0114] Specifically, an automated control system and conveyor system management software can be used for regulation. First, the system receives the classification planning path and defect label information of the rock wool curtain wall panels. For example, assume a rock wool curtain wall panel is marked with a "severe" defect. The system determines the required conveying speed according to the preset rules. The setting rules are as follows: for rock wool curtain wall panels with "severe" defects, the conveying speed should be set to 0.5 m / s to ensure rapid processing; while for rock wool curtain wall panels with "minor" defects, the conveying speed can be set to 1.0 m / s. Input these rules into the control system and combine them with the current state of the conveyor belt (such as load condition, length of the conveyor belt, etc.). The control system monitors the operating state of the conveyor belt in real time and dynamically adjusts the conveying speed according to the classification information of the rock wool curtain wall panels. For example, if the system detects that the current conveyor belt is heavily loaded, it automatically reduces the conveying speed to avoid overloading. Finally, the system generates and records the conveying rate parameters of each rock wool curtain wall panel.

[0115] Step S454: Construct a three-dimensional model of the flexible conveyor system to obtain a conveying execution collaborative control model;

[0116] Specifically, computer-aided design (CAD) software, such as AutoCAD or SolidWorks, can be used to construct a three-dimensional model of the flexible conveyor system. First, collect all component information of the flexible conveyor system, including the dimensions and specifications of the conveyor belt, drive device, sensors, and control unit, etc. Create a three-dimensional model of the conveyor system in the CAD software and gradually add each component. For example, first draw the path of the conveyor belt, then add the drive motor and support structure, and finally integrate the sensors and control unit. In this way, the layout and structure of the entire conveyor system can be accurately reflected in the virtual environment. Set the connection relationships and working parameters between each component in the model. For example, the rotational speed of the drive motor, the load capacity of the conveyor belt, and the monitoring range of the sensors. Through these settings, the performance of the conveyor system during actual operation can be simulated. Finally, import the constructed three-dimensional model into the control system to form a conveying execution collaborative control model.

[0117] Step S455: Based on the conveying execution collaborative control model, perform a transportation simulation on the rock wool curtain wall panels according to the spatial dynamic coordinates of the rock wool curtain wall panels, the grasping path of the rock wool curtain wall panels, and the conveying rate parameters of the rock wool curtain wall panels to obtain the transportation simulation process data of the rock wool curtain wall panels;

[0118] Specifically, professional logistics simulation software such as FlexSim or Arena can be used to simulate the transportation process of rock wool curtain wall panels. Import the conveying execution collaborative control model into the simulation software and set the model parameters, including the spatial dynamic coordinates, grasping path, and conveying rate parameters of the rock wool curtain wall panels. According to the actual spatial dynamic coordinates of the rock wool curtain wall panels, set the initial position and moving trajectory in the simulation software. According to the grasping path of the rock wool curtain wall panels, define the path nodes and path directions in the simulation. According to the conveying rate parameters of the rock wool curtain wall panels, set the conveyor belt speed and acceleration in the simulation. In the simulation software, simulate the entire transportation process of the rock wool curtain wall panels from the production line to the storage area. The software will dynamically display the movement of the rock wool curtain wall panels on the conveyor belt according to the input parameters and model, and record the simulation data of each step, including the position, speed, and acceleration of the rock wool curtain wall panels, etc. These data will be collected and organized into the rock wool curtain wall panel transportation simulation process data.

[0119] Step S456: Predict the arrival time of the rock wool curtain wall panels based on the rock wool curtain wall panel transportation simulation process data to obtain the predicted storage arrival time of the rock wool curtain wall panels; use the predicted storage arrival time of the rock wool curtain wall panels as the time point when the rock wool curtain wall panels are grabbed on the transportation line, and record the grasping path of the rock wool curtain wall panels, the conveying rate parameters of the rock wool curtain wall panels, and the time point when the rock wool curtain wall panels are grabbed on the transportation line as the classified conveying plan of the rock wool curtain wall panels.

[0120] Specifically, data analysis software such as the Pandas library of Python or the R language can be used to process the rock wool curtain wall panel transportation simulation process data and perform arrival time prediction. Analyze the simulation data to find out the speed changes and path characteristics of the rock wool curtain wall panels during the transportation process. According to the conveying rate parameters and path characteristics of the rock wool curtain wall panels, use statistical or machine learning methods to predict the time when the rock wool curtain wall panels reach the storage area. For example, establish a linear regression model, take the conveying rate and path length of the rock wool curtain wall panels as independent variables, and the arrival time as the dependent variable, and predict the arrival time of the rock wool curtain wall panels through the model. After the prediction is completed, record the predicted arrival time as the predicted storage arrival time of the rock wool curtain wall panels. Use this time point as the time point when the rock wool curtain wall panels are grabbed on the transportation line. Integrate the grasping path, conveying rate parameters, and the grabbed time point of the rock wool curtain wall panels to form the classified conveying plan of the rock wool curtain wall panels. For example, if a rock wool curtain wall panel is expected to reach the storage area in 5 minutes, record its predicted time as 5 minutes and note in the plan: starting from the production line, conveying at a speed of 0.8 m / s, and expected to reach the No. 2 storage area in 5 minutes, finally obtaining a detailed classified conveying plan of the rock wool curtain wall panels.

[0121] The present invention improves the positioning accuracy and operation efficiency of rock wool curtain wall panels through real-time three-dimensional coordinate tracking, ensuring the accuracy of grasping and placing. The grasping path is optimized through path planning under dynamic coordinate guidance, reducing unnecessary movement and time waste. By dynamically adjusting the conveying speed according to the characteristics of rock wool curtain wall panels, the flexibility of the conveying process is increased, and the requirements of different panels are met. By establishing a collaborative control model for three-dimensional models and conveying execution, the collaborative control ability of the system is enhanced. By transportation simulation and arrival time prediction, the accuracy of simulation and prediction is improved, optimizing logistics scheduling and planning. By adjusting the system in real time to respond to production changes, the response speed and flexibility of the production line are improved, reducing machine waiting time, thereby improving the overall efficiency.

[0122] Preferably, step S5 includes the following steps:

[0123] Step S51: Based on a preset detection record database of rock wool curtain wall panels, perform detection data association on the storage coordinates of classified rock wool curtain wall panels to obtain a detection traceability link for rock wool curtain wall panels;

[0124] Specifically, a database management system such as MySQL or Oracle can be used to store and manage the detection records of rock wool curtain wall panels. These records include the detection results, storage locations, and other relevant information of each rock wool curtain wall panel. Assign a unique identification code (ID) to each rock wool curtain wall panel, and associate this ID with its storage coordinates. When a rock wool curtain wall panel is stored, use a handheld scanner to scan its ID and input the scan result into the database. The query system in the database will retrieve all the detection data related to this rock wool curtain wall panel according to this ID, including information such as its defect level, detection time, and detection personnel. Use data association techniques such as SQL queries or Python scripts to match the storage coordinates of rock wool curtain wall panels with the detection data. For example, write an SQL query to join the coordinates in the storage location table with the ID in the detection result table to obtain the detection traceability link for each rock wool curtain wall panel. Finally, organize these associated data into a traceability link, which details the entire process of each rock wool curtain wall panel from production to storage.

[0125] Step S52: Backtrack and collect the production environment data of the rock wool curtain wall panels according to the detection traceability link of the rock wool curtain wall panels to obtain a set of historical production environment parameters;

[0126] Specifically, a production management system (such as an MES system) and a data acquisition system (such as a SCADA system) can be used to collect the production environment data of rock wool curtain wall panels. These systems can monitor various parameters on the production line in real time, such as temperature, humidity, pressure, etc. According to the detection and traceability link of rock wool curtain wall panels, determine the batches of rock wool curtain wall panels for which data needs to be retrospectively collected. Use the historical data query function of the SCADA system to query the environmental parameter records of these batches during the production process. For example, query the temperature and humidity data of the production line within a specific time period, and these data correspond to the production batches and times of the rock wool curtain wall panels. Export these environmental parameter data and match them with the detection data of the rock wool curtain wall panels. For example, associate the detection results of each rock wool curtain wall panel with the environmental parameters during its production to form a set of historical production environmental parameters.

[0127] Step S53: Based on the detection and traceability link of rock wool curtain wall panels and the set of historical production environmental parameters, digitally reconstruct the production process to obtain a production quality management model;

[0128] Specifically, data analysis and modeling software, such as Python combined with Pandas, NumPy, and Scikit - learn libraries, or professional statistical software such as SAS or SPSS, can be used to process the data of the detection and traceability link of rock wool curtain wall panels and the set of historical production environmental parameters. Extract the data of the detection and traceability link of rock wool curtain wall panels from the database, including information such as the detection results, storage locations, and defect labels of each rock wool curtain wall panel. At the same time, extract the corresponding historical production environmental parameters from the production management system, such as temperature, humidity, raw material batches, etc. Use these data to construct a production quality management model. For example, construct a multiple regression model, taking the defect label of the rock wool curtain wall panel as the dependent variable and the production environmental parameters as the independent variables, and analyze the influence of different environmental parameters on the quality of the rock wool curtain wall panel. Through model analysis, identify the key factors affecting the quality of the rock wool curtain wall panel and quantify the degree of influence of these factors on the quality. Use this model as the production quality management model to predict and monitor the quality changes during the production process.

[0129] Step S54: Trace the defects of the production quality management model according to the defect labels of the rock wool curtain wall panels to obtain the evaluation data of the defect source links, and generate optimization suggestions for the production process of the rock wool curtain wall panels based on the evaluation data of the defect source links to obtain a production suggestion report for the rock wool curtain wall panels;

[0130] Specifically, data analysis software and report generation tools, such as Tableau or Power BI, can be used to conduct defect traceability analysis on the production quality management model and generate an optimization suggestion report. According to the defect labels of the rock wool curtain wall panels, such as "severe", "medium", and "slight", identify the production environment parameters related to these defects in the production quality management model. For example, it is found that rock wool curtain wall panels produced under high humidity conditions are more likely to have "severe" defects. Conduct in-depth analysis of this data to identify the source links of defects in the production process. For example, use clustering analysis to identify batches of rock wool curtain wall panels with similar defect characteristics and analyze the production environment parameters of these batches to find the common source of defects. Finally, generate a production suggestion report for rock wool curtain wall panels based on the evaluation data of the defect source links. The report includes the results of defect traceability and targeted optimization suggestions, such as adjusting production parameters, improving raw material quality control, or optimizing the production process. For example, if it is found that high humidity is the main cause of defects, it is recommended to increase dehumidification measures under high humidity conditions or adjust the production plan to avoid production under high humidity conditions, and finally obtain a production suggestion report for rock wool curtain wall panels.

[0131] Step S55: Apply the production suggestion report for rock wool curtain wall panels to the rock wool curtain wall panel production line and execute Steps S1 - S3 for the newly produced rock wool curtain wall panels, so as to obtain new defect labels for the rock wool curtain wall panels.

[0132] Specifically, the improvement measures proposed in the production suggestion report can be discussed with the person in charge of the production line and engineers, and a specific implementation plan can be formulated. For example, the report recommends optimizing in aspects such as humidity control, raw material screening, and process parameter adjustment. According to these suggestions, update the operation guidelines and monitoring system settings of the production line. On the production line, install or adjust humidity control equipment to ensure that the humidity of the production environment is maintained within the optimal range. Conduct more stringent quality control over the raw material supply, such as increasing the inspection frequency and standards of raw materials. Adjust process parameters, such as adjusting temperature and pressure settings, to reduce the occurrence of defects. Continue to execute Steps S1 - S3 for the newly produced rock wool curtain wall panels. After executing these steps, finally obtain the defect labels of the new rock wool curtain wall panels. For example, if it is found that the number of "severe" defects in the newly produced rock wool curtain wall panels has decreased significantly while the number of "slight" defects has increased, this indicates that the improvement measures have achieved certain results.

[0133] The present invention significantly enhances the quality traceability ability of rock wool curtain wall panels by establishing a detection data association and traceability link, and can effectively track the product quality status. The data utilization rate is improved by collecting historical production environment parameters. Through the digital reconstruction of the production process and the creation of a quality management model, the digitalization of production management is realized, and the management efficiency and accuracy are improved. The defect source is identified through the defect traceability function, providing a basis for preventive measures and process optimization. Through the production suggestion report generated based on the defect assessment data, a targeted optimization plan is provided, which helps to improve the product quality. Applying these suggestions and repeatedly executing the detection steps forms a continuous quality improvement mechanism. The establishment of the production quality management model improves the production transparency, enabling managers to intuitively understand the production status.

[0134] Preferably, the present invention also provides a defect detection system for rock wool curtain wall panels based on machine vision, which is used to execute the defect detection method for rock wool curtain wall panels based on machine vision as described above. The defect detection system for rock wool curtain wall panels based on machine vision includes:

[0135] An image acquisition module, which is used to perform multi-spectral image acquisition on the surface of the rock wool curtain wall panel to obtain a multi-spectral image set of the rock wool curtain wall panel; and perform feature extraction on the multi-spectral image set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel.

[0136] A defect extraction module, which is used to perform geometric defect segmentation on the initial feature map of the rock wool curtain wall panel to obtain a geometric defect feature map of the rock wool curtain wall panel; perform material anomaly division on the initial feature map of the rock wool curtain wall panel to obtain a material defect feature map of the rock wool curtain wall panel; and perform feature alignment and fusion on the geometric defect feature map and the material defect feature map of the rock wool curtain wall panel to obtain a comprehensive defect feature map of the rock wool curtain wall panel.

[0137] A defect classification module, which is used to calculate defect quantification indexes for the comprehensive defect feature map of the rock wool curtain wall panel to obtain a defect feature vector of the rock wool curtain wall panel; and perform defect classification mapping on the defect feature vector of the rock wool curtain wall panel to obtain a defect label of the rock wool curtain wall panel.

[0138] A classification control module, which is used to perform intelligent classification guidance on the rock wool curtain wall panel according to the defect label of the rock wool curtain wall panel to obtain a classification planning path of the rock wool curtain wall panel; perform collaborative control on the flexible conveying system for the classification planning path of the rock wool curtain wall panel to obtain a classification conveying plan of the rock wool curtain wall panel; and position the rock wool curtain wall panel according to the classification conveying plan of the rock wool curtain wall panel to obtain the classification storage coordinates of the rock wool curtain wall panel.

[0139] The traceability optimization module is used to detect data association for the storage coordinates of classified rock wool curtain wall panels to obtain the detection traceability link of rock wool curtain wall panels; digitally reconstruct the production process based on the detection traceability link to obtain a production quality management model; generate optimization suggestions for the production process of rock wool curtain wall panels according to the production quality management model to obtain a production suggestion report for rock wool curtain wall panels; apply the production suggestion report for rock wool curtain wall panels to the rock wool curtain wall panel production line, and execute Steps S1 - S3 for the newly produced rock wool curtain wall panels, thereby obtaining defect labels for the new rock wool curtain wall panels.

[0140] Therefore, from any perspective, the embodiments should be regarded as exemplary and non - restrictive. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0141] The above - mentioned are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting defects of rock wool curtain wall panels based on machine vision, characterized in that: The following steps are involved: Step S1: collecting multispectral images of the surface of the rock wool curtain wall panel to obtain a multispectral image set of the rock wool curtain wall panel; extracting features from the multispectral image set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel; Step S2: performing geometric defect segmentation on the initial feature map of the rock wool curtain wall panel to obtain a geometric defect feature map of the rock wool curtain wall panel; performing material anomaly segmentation on the initial feature map of the rock wool curtain wall panel to obtain a material defect feature map of the rock wool curtain wall panel; performing feature alignment and fusion on the geometric defect feature map of the rock wool curtain wall panel and the material defect feature map of the rock wool curtain wall panel to obtain a comprehensive defect feature map of the rock wool curtain wall panel; Step S3: Calculate defect quantification indexes on the comprehensive defect feature map of the rock wool curtain wall panel to obtain a defect feature vector of the rock wool curtain wall panel; perform defect classification mapping on the defect feature vector of the rock wool curtain wall panel to obtain a defect label of the rock wool curtain wall panel; Step S4: intelligently classify and guide the rock wool curtain wall panels according to the defect labels of the rock wool curtain wall panels to obtain a classification planning path for the rock wool curtain wall panels; perform collaborative control of the flexible conveying system on the classification planning path for the rock wool curtain wall panels to obtain a classification conveying plan for the rock wool curtain wall panels; locate the rock wool curtain wall panels according to the classification conveying plan for the rock wool curtain wall panels to obtain classification storage coordinates for the rock wool curtain wall panels; Step S5: associate the detection data of the stored coordinates of the classified rock wool curtain wall panels to obtain the rock wool curtain wall panel detection traceability link; digitally reconstruct the production process based on the detection traceability link to obtain a production quality management model; generate optimization suggestions for the rock wool curtain wall panel production process according to the production quality management model to obtain a rock wool curtain wall panel production suggestion report; apply the rock wool curtain wall panel production suggestion report to the rock wool curtain wall panel production line, and execute steps S1-S3 on the newly produced rock wool curtain wall panels to obtain defect labels for new rock wool curtain wall panels.

2. The machine vision-based rock wool curtain wall panel defect detection method according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: collecting multispectral images of the surface of the rock wool curtain wall panel to obtain an initial multispectral image set of the rock wool curtain wall panel; Step S12: performing spectral pre-calibration on the initial rock wool curtain wall panel multispectral image set to obtain a rock wool curtain wall panel multispectral image set; Step S13: performing frequency domain enhancement on the multispectral image set of the rock wool curtain wall panel to obtain a frequency domain enhanced multispectral image set; Step S14: performing spatial domain enhancement on the frequency domain enhanced multispectral image set to obtain a frequency domain-spatial domain enhanced multispectral image set; Step S15: extract high-dimensional features from the frequency-domain-spatial-domain enhanced multispectral image set to obtain a high-dimensional feature set of the rock wool curtain wall panel, and perform multimodal feature fusion on the high-dimensional feature set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel.

3. The machine vision-based rock wool curtain wall panel defect detection method according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: extracting geometric shape features from the initial feature map of the rock wool curtain wall panel to obtain a geometric feature descriptor of the rock wool curtain wall panel; Step S22: performing defect edge detection on the geometric feature descriptor of the rock wool curtain wall panel to obtain the geometric defect contour of the rock wool curtain wall panel, and performing defect connectivity identification on the geometric defect contour of the rock wool curtain wall panel to obtain defect network connectivity data; Step S23: performing defect morphological dimension identification on the geometric defect contour of the rock wool curtain wall panel to obtain defect geometric dimension distribution data, and performing defect space entropy density calculation on the geometric defect contour of the rock wool curtain wall panel to obtain defect space distribution complexity data; Step S24: performing feature clustering on the geometric defect contour of the rock wool curtain wall panel according to the defect network connectivity data, the defect geometric dimension distribution data and the defect spatial distribution complexity data to obtain the geometric defect pattern classification data of the rock wool curtain wall panel, and annotating the geometric defect characteristics of the initial feature map of the rock wool curtain wall panel according to the rock wool curtain wall panel geometric defect pattern classification data to obtain the rock wool curtain wall panel geometric defect feature map; Step S25: performing material anomaly classification on the initial feature map of the rock wool curtain wall panel to obtain a material defect feature map of the rock wool curtain wall panel; Step S26: performing defect feature weighted fusion on the rock wool curtain wall panel geometric defect feature map and the rock wool curtain wall panel material defect feature map to obtain a comprehensive defect feature map of the rock wool curtain wall panel.

4. The machine vision-based rock wool curtain wall panel defect detection method according to claim 3 is characterized in that: Step S25 includes the following steps: Step S251: extracting material texture features from the initial feature map of the rock wool curtain wall panel to obtain material feature mapping data of the rock wool curtain wall panel; Step S252: performing material anomaly identification on the material feature mapping data of the rock wool curtain wall panel to obtain the material anomaly mapping data of the rock wool curtain wall panel, and performing material feature gradient calculation on the initial feature map of the rock wool curtain wall panel according to the material anomaly mapping data of the rock wool curtain wall panel to obtain the material feature gradient of the rock wool curtain wall panel; Step S253: quantifying the abnormal strength of the characteristic gradient of the rock wool curtain wall panel material to obtain the defect strength data of the rock wool curtain wall panel material; Step S254: performing material defect classification mapping on the rock wool curtain wall panel material defect intensity data to obtain a rock wool curtain wall panel material defect label; Step S255: annotating the material characteristics of the initial feature map of the rock wool curtain wall panel according to the material defect label of the rock wool curtain wall panel to obtain the material defect feature map of the rock wool curtain wall panel.

5. The machine vision-based rock wool curtain wall panel defect detection method according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: performing pixel-level defect area statistics on the comprehensive defect feature map of the rock wool curtain wall panel to obtain the defect coverage area of ​​the rock wool curtain wall panel; Step S32: extracting morphological defect depth features from the comprehensive defect feature map of the rock wool curtain wall panel to obtain defect depth distribution data of the rock wool curtain wall panel; Step S33: normalizing the scale of the rock wool curtain wall panel defect depth distribution data to obtain a rock wool curtain wall panel defect depth index set; Step S34: performing edge feature topological complexity evaluation on the comprehensive feature graph of rock wool curtain wall panel defects to obtain a defect edge complexity evaluation data set; Step S35: Calculate defect quantification indexes for the rock wool curtain wall panel defect comprehensive feature map according to the rock wool curtain wall panel defect coverage area, the rock wool curtain wall panel defect depth index set and the defect edge complexity evaluation data set to obtain a rock wool curtain wall panel defect feature vector; Step S36: performing defect classification mapping on the rock wool curtain wall panel defect feature vector according to preset evaluation rules to obtain a rock wool curtain wall panel defect label.

6. The machine vision-based rock wool curtain wall panel defect detection method according to claim 5 is characterized in that: Step S35 includes the following steps: Step S351: Calculate the standard deviation of the rock wool curtain wall panel defect depth index set and the defect edge complexity evaluation data set to obtain the defect depth standard deviation and the defect edge complexity standard deviation; Step S352: Calculate the mean of the defect depth index set and the defect edge complexity evaluation data set of the rock wool curtain wall panel respectively to obtain the defect depth average value and the defect edge complexity average value; Step S353: Calculate the coefficient of variation of the defect depth index set of the rock wool curtain wall panel according to the defect depth standard deviation and the defect depth average value to obtain the defect depth topological coefficient of variation; Step S354: Calculate the coefficient of variation of the rock wool curtain wall panel defect depth index set according to the defect edge complexity standard deviation and the defect edge complexity average value to obtain the defect edge topological variation coefficient, and record the rock wool curtain wall panel defect coverage area, the sink depth topological variation coefficient and the defect depth topological variation coefficient as the rock wool curtain wall panel defect feature vector.

7. The machine vision-based rock wool curtain wall panel defect detection method according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: collecting a transportation route distribution map of the rock wool curtain wall panels to obtain a transportation route network topology map; Step S42: performing data preprocessing on the transport route network topology map to obtain a transport path feature vector, wherein the data preprocessing includes path node weight calculation, path connection feature extraction and transport path capacity evaluation; Step S43: constructing graded transportation constraint conditions for the rock wool curtain wall panels according to the defect labels of the rock wool curtain wall panels to obtain graded transportation path selection criteria; Step S44: performing target transportation path optimization calculation on the rock wool curtain wall panels based on the graded transportation path selection criteria and the transportation path feature vector to obtain a classified planning path for the rock wool curtain wall panels; Step S45: using the flexible conveying system end effector to perform coordinated control of the rock wool curtain wall panel classification planning path to obtain a rock wool curtain wall panel classification conveying plan; Step S46: the rock wool curtain wall panels are automatically transported according to the rock wool curtain wall panel classification and transportation plan, and the rock wool curtain wall panels are positioned after the operation is completed to obtain the classification storage coordinates of the rock wool curtain wall panels.

8. The machine vision-based rock wool curtain wall panel defect detection method according to claim 7 is characterized in that: Step S45 includes the following steps: Step S451: performing real-time three-dimensional coordinate tracking on the rock wool curtain wall panel to obtain the spatial dynamic coordinates of the rock wool curtain wall panel; Step S452: planning a grabbing path for the rock wool curtain wall panel according to the spatial dynamic coordinates of the rock wool curtain wall panel to obtain a grabbing path for the rock wool curtain wall panel; Step S453: regulating the conveyor belt speed of the flexible conveying system according to the rock wool curtain wall panel classification planning path and the rock wool curtain wall panel defect label to obtain the rock wool curtain wall panel conveying rate parameter; Step S454: constructing a three-dimensional model of the flexible conveying system to obtain a conveying execution collaborative control model; Step S455: Based on the transportation execution collaborative control model, the rock wool curtain wall panel is transported according to the spatial dynamic coordinates of the rock wool curtain wall panel, the rock wool curtain wall panel grabbing path and the rock wool curtain wall panel transportation rate parameters to obtain the rock wool curtain wall panel transportation simulation process data; Step S456: Predict the arrival time of the rock wool curtain wall panels based on the rock wool curtain wall panel transportation simulation process data to obtain the predicted storage arrival time of the rock wool curtain wall panels; use the predicted storage arrival time of the rock wool curtain wall panels as the time point when the rock wool curtain wall panels are picked up on the transportation line, and record the rock wool curtain wall panel grabbing path, rock wool curtain wall panel transportation rate parameters and the time point when the rock wool curtain wall panels are picked up on the transportation line as the rock wool curtain wall panel classification transportation plan.

9. The machine vision-based rock wool curtain wall panel defect detection method according to claim 1, characterized in that: Step S5 includes the following steps: Step S51: Based on a preset rock wool curtain wall panel detection record database, detection data association is performed on the storage coordinates of the classified rock wool curtain wall panels to obtain a rock wool curtain wall panel detection traceability link; Step S52: According to the rock wool curtain wall panel detection traceability link, the production environment data of the rock wool curtain wall panel is retrospectively collected to obtain a historical production environment parameter set; Step S53: digitally reconstruct the production process based on the rock wool curtain wall panel detection traceability link and the historical production environment parameter set to obtain a production quality management model; Step S54: trace the defect source of the production quality management model according to the defect label of the rock wool curtain wall panel to obtain the defect source link evaluation data, and generate optimization suggestions for the rock wool curtain wall panel production process according to the defect source link evaluation data to obtain the rock wool curtain wall panel production suggestion report; Step S55: Apply the rock wool curtain wall panel production suggestion report to the rock wool curtain wall panel production line, and perform steps S1 to S3 on the newly produced rock wool curtain wall panels, thereby obtaining defect labels of the new rock wool curtain wall panels.

10. A machine vision-based rock wool curtain wall panel defect detection system, characterized in that: Used to execute the rock wool curtain wall panel defect detection method based on machine vision as claimed in claim 1, the rock wool curtain wall panel defect detection system based on machine vision comprises: An image acquisition module is used to acquire multispectral images of the surface of the rock wool curtain wall panel to obtain a multispectral image set of the rock wool curtain wall panel; and to extract features from the multispectral image set of the rock wool curtain wall panel to obtain an initial feature map of the rock wool curtain wall panel; The defect extraction module is used to perform geometric defect segmentation on the initial feature map of the rock wool curtain wall panel to obtain the geometric defect feature map of the rock wool curtain wall panel; perform material anomaly segmentation on the initial feature map of the rock wool curtain wall panel to obtain the material defect feature map of the rock wool curtain wall panel; perform feature alignment and fusion on the geometric defect feature map of the rock wool curtain wall panel and the material defect feature map of the rock wool curtain wall panel to obtain the comprehensive defect feature map of the rock wool curtain wall panel; The defect classification module is used to calculate the defect quantification index of the comprehensive defect feature map of the rock wool curtain wall panel to obtain the defect feature vector of the rock wool curtain wall panel; perform defect classification mapping on the defect feature vector of the rock wool curtain wall panel to obtain the defect label of the rock wool curtain wall panel; The classification control module is used to intelligently classify and guide the rock wool curtain wall panels according to the defect labels of the rock wool curtain wall panels to obtain the classification planning path of the rock wool curtain wall panels; to coordinate the flexible conveying system on the classification planning path of the rock wool curtain wall panels to obtain the classification conveying plan of the rock wool curtain wall panels; to locate the rock wool curtain wall panels according to the classification conveying plan of the rock wool curtain wall panels to obtain the classification storage coordinates of the rock wool curtain wall panels; The traceability optimization module is used to associate the detection data of the storage coordinates of the classified rock wool curtain wall panels to obtain the rock wool curtain wall panel detection traceability link; digitally reconstruct the production process based on the detection traceability link to obtain the production quality management model; generate optimization suggestions for the rock wool curtain wall panel production process according to the production quality management model to obtain a rock wool curtain wall panel production suggestion report; apply the rock wool curtain wall panel production suggestion report to the rock wool curtain wall panel production line, and execute steps S1 to S3 on the newly produced rock wool curtain wall panels to obtain the defect labels of the new rock wool curtain wall panels.

Images