Method and system for detecting quality of rock wool board

Abstract

The invention relates to the technical field of building material detection, in particular to a rock wool board quality detection method and system, and the method comprises the steps: shooting the surface of a rock wool board through an industrial camera, and obtaining the surface image data of the rock wool board; carrying out data preprocessing on the obtained rock wool board surface image data, wherein the data preprocessing comprises graying processing and filtering denoising; after data preprocessing is completed, edge features and texture features of the rock wool board surface image data are extracted; after feature extraction is completed, a rock wool board quality detection model based on a convolutional neural network and a support vector machine is constructed and trained; after model training is completed, rock wool board quality detection is conducted, and a detection result is output. According to the method, the features of the rock wool board are captured through image data processing in combination with the convolutional neural network and the support vector machine to train the rock wool board quality detection model, the quality condition of the rock wool board is accurately detected, and the rock wool board quality detection efficiency and the automation level are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of building material testing, and in particular to a method and system for quality inspection of rock wool boards. Background Art

[0002] Rock wool boards are widely used in the fields of building insulation and the like. Their appearance integrity and quality have an important impact on the service performance and aesthetics. The traditional method of relying on manual visual inspection of the appearance and quality of rock wool boards has many drawbacks. On the one hand, the inspection efficiency is low. Facing large-scale production tasks, manual visual inspection is time-consuming and laborious, and it is difficult to meet the fast production rhythm. On the other hand, the accuracy of the inspection results depends on the professional qualities and subjective judgments of the inspectors. There are differences in the understanding and judgment criteria of defects among different personnel, resulting in frequent misjudgments and missed judgments. For example, for defects such as fine fiber exposure, small holes, and unobvious delamination, inexperienced inspectors may overlook them and fail to detect the defects. Due to the nature of subjective judgment, the quality assessment lacks consistency and cannot meet the requirements of large-scale high-quality production. With the development of technology, there is an urgent need for an intelligent and high-precision method to detect the quality of rock wool boards. Summary of the Invention

[0003] In view of the above-mentioned technical deficiencies, the present invention provides a method and system for quality inspection of rock wool boards.

[0004] The present invention is achieved through the following technical solutions:

[0005] A method for quality inspection of rock wool boards is provided. The method includes the following steps:

[0006] Step S10: Obtain the surface image data of the rock wool board by shooting the surface of the rock wool board with an industrial camera;

[0007] Step S20: Perform data preprocessing on the obtained surface image data of the rock wool board, including grayscale processing and filtering denoising, to obtain the grayscale image of the rock wool board surface;

[0008] Step S30: Extract the texture features of the obtained grayscale image of the rock wool board surface to obtain the grayscale image of the rock wool board surface with the texture features extracted;

[0009] Step S40: Collect the grayscale image samples of normal rock wool boards without defects and the grayscale image samples of rock wool boards with various quality defects, and construct and train a rock wool board quality detection model based on a convolutional neural network and a support vector machine;

[0010] Step S50: After the model training is completed, input the grayscale image of the rock wool board surface with the texture features extracted in Step S30 into the model for quality inspection of the rock wool board and output the inspection result;

[0011] Among them, the industrial camera used in step S10 has a 2k resolution. A ring light source is installed around the industrial camera, and the rock wool board is photographed from the front and side to obtain surface image data of the rock wool board at different angles.

[0012] Among them, the data preprocessing steps in step S20 include grayscale processing and filtering denoising. Grayscale processing is used to simplify the data and display the texture features of the rock wool board surface; filtering denoising is used to remove noise interference and retain the edge features and texture features of the rock wool board.

[0013] Among them, when outputting the detection result in step S50, the position where the defect exists and the information of the defect type are marked on the surface image of the rock wool board through visualization software, and a detection report is generated; at the same time, according to the preset quality standard of the rock wool board, when the number and severity of the detected defects are within the range of the quality standard of the rock wool board, it is determined that the quality of the rock wool board is qualified; when the number and severity of the detected defects exceed the range of the quality standard of the rock wool board, it is determined that the quality of the rock wool board is unqualified and targeted improvement suggestions are generated.

[0014] Preferably, the steps of data preprocessing the obtained grayscale image of the rock wool board surface in step S20 include:

[0015] Grayscale processing: Using the weighted average method, for the obtained surface image data of the rock wool board, read the information of the red, green, and blue color channels of each pixel point in the image, and calculate the grayscale value of each pixel point, as shown in Equation (1):

[0016] Gray = 0.299 * R + 0.587 * G + 0.114 * B (1)

[0017] Among them, Gray is the grayscale value obtained after weighted average calculation, and this value is also the pixel value of the corresponding pixel point in the grayscale image. R is the color value of the red channel in the surface image data of the rock wool board, which reflects the brightness information of this pixel point in the red spectrum. G is the color value of the green channel in the surface image data of the rock wool board, which reflects the brightness information of this pixel point in the green spectrum. B is the color value of the blue channel in the surface image data of the rock wool board, which reflects the brightness information of this pixel point in the blue spectrum; replace the three color channel values of the original pixel point with the calculated grayscale value to obtain a single-channel grayscale image of the rock wool board surface.

[0018] Filtering denoising: For each pixel point (x, y) in the obtained grayscale image of the rock wool board surface, determine a rectangular neighborhood, with a size of 3×3, 5×5, etc. Then the pixel points in the rectangular neighborhood are (x1, y1), (x2, y2),..., (x n , y n ), and the corresponding grayscale values are Gray1, Gray2,..., Grayn , where n represents that there are n pixel points in the rectangular neighborhood, and the spatial Gaussian weight ω is calculated s (x, y, x k , y k ), as shown in Equation (2):

[0019]

[0020] Among them, (x, y) is the current pixel point in the rectangular neighborhood, and (x k , y k ) is the k-th pixel point in the rectangular neighborhood, 0 < k ≤ n, and σ s is the spatial standard deviation, which represents the attenuation rate of the spatial weight and is used to measure the spatial distance relationship between the pixel points in the rectangular neighborhood and the current pixel point (x, y). The closer the distance, the greater the weight; at the same time, the gray-scale Gaussian weight ω r (g(x, y), g(x k , y k )) is calculated, as shown in Equation (3):

[0021]

[0022] Among them, g(x, y) is the gray value of the current pixel point in the rectangular neighborhood, and g(x k , y k ) is the gray value corresponding to the k-th pixel point in the rectangular neighborhood, and σ r is the gray-scale standard deviation, which determines the change speed of the gray-scale weight and is used to measure the similarity of the gray values between the pixel points in the rectangular neighborhood and the current pixel point (x, y). The closer the gray values, the greater the weight; after obtaining the spatial standard deviation and the gray-scale standard deviation, multiply the spatial standard deviation and the gray-scale standard deviation to obtain the filtering weight ω(x, y, x k , y k ), and the filtered gray value Gray’(x, y) is obtained through calculation, as shown in Equation (4):

[0023]

[0024] Traverse each point in the gray-scale image of the rock wool board surface, and replace the gray value of the pixel point in the original gray-scale image of the rock wool board surface with Gray’(x, y) to obtain the gray-scale image of the rock wool board surface after filtering and denoising.

[0025] Preferably, in step S30, feature extraction is performed on the texture features of the obtained gray-scale image of the rock wool board surface. By using the method of constructing a gray-level co-occurrence matrix, it includes:

[0026] Construct the gray-level co-occurrence matrix in the 0° direction: Constructing the gray-level co-occurrence matrix in the 0° direction means that when constructing the gray-level co-occurrence matrix centered on a certain pixel point (x, y) in the gray-scale image of the surface of the rock wool board, the frequency of the gray-level value combinations of the adjacent pixel point (x + 1, y) directly to the right of this point is counted to construct the gray-level co-occurrence matrix in the 0° direction; The specific operation is to read the gray-level value of the pixel point (x, y), set it as i, and at the same time read the gray-level value of its adjacent pixel point to the right (x + 1, y), set it as j. In a two-dimensional matrix with all initial values of 0, that is, the gray-level co-occurrence matrix to be constructed in the 0° direction, find the corresponding positions (the i-th row, j-th column and the j-th row, i-th column, the gray-level co-occurrence matrix is symmetric), and add 1 to the element value at this position. Traverse all pixel points of the entire gray-scale image of the surface of the rock wool board in the order from left to right and from top to bottom to obtain the gray-level co-occurrence matrix in the 0° direction; This direction can capture the texture change characteristics on the horizontal direction of the surface of the rock wool board. For example, for some rock wool boards with relatively regular fiber arrangements, the distribution of gray-level value combinations in the 0° direction is relatively regular, while when there are defects such as horizontal cracks or holes, the distribution of gray-level value combinations will appear abnormal;

[0027] Construct the gray-level co-occurrence matrix in the 45° direction: Constructing the gray-level co-occurrence matrix in the 45° direction means that when constructing the gray-level co-occurrence matrix centered on a certain pixel point (x, y) in the gray-scale image of the surface of the rock wool board, the frequency of the gray-level value combinations of the adjacent pixel point (x + 1, y + 1) at a 45° direction to the upper right of this point is counted to construct the gray-level co-occurrence matrix in the 45° direction; Read the gray-level value of the pixel point (x, y) and the gray-level value of the adjacent pixel point (x + 1, y + 1) at a 45° direction to the upper right of this point. According to the same operation method above, traverse all pixel points of the entire gray-scale image of the surface of the rock wool board to obtain the gray-level co-occurrence matrix in the 45° direction; This direction can capture the texture change characteristics on the 45° direction to the upper right of the surface of the rock wool board. For example, when there are oblique scratches on the surface of the rock wool board or there is a special distribution trend of fibers in this direction due to the production process, the gray-level co-occurrence matrix in the 45° direction can reflect the change of gray-level value combinations, and then assist in judging the surface defect situation of the rock wool board;

[0028] Construct the gray-level co-occurrence matrix in the 90° direction: The 90° direction means that when constructing the gray-level co-occurrence matrix with a certain pixel point (x, y) in the gray-scale image of the rock wool board surface as the center, the frequency of the gray-level value combination of the adjacent pixel point (x, y + 1) in the vertical direction of this point is statistically counted to construct the gray-level co-occurrence matrix in the 90° direction; read the gray-level value of the pixel point (x, y) and the gray-level value of the adjacent pixel point (x, y + 1) in the vertical direction of this point, and according to the same operation method above, traverse all pixel points of the entire gray-scale image of the rock wool board surface to obtain the gray-level co-occurrence matrix in the 90° direction; this direction can capture the texture change characteristics on the vertical direction of the rock wool board surface. For example, when there are delamination defects on the rock wool board, there are obvious hierarchical changes in the vertical direction, and the gray-level co-occurrence matrix in the 90° direction will show a probability distribution of gray-level value combinations that is completely different from that of the normal area, which helps to accurately identify such defects;

[0029] Construct the gray-level co-occurrence matrix in the 135° direction: The 135° direction means that when constructing the gray-level co-occurrence matrix with a certain pixel point (x, y) in the gray-scale image of the rock wool board surface as the center, the frequency of the gray-level value combination of the adjacent pixel point (x - 1, y - 1) in the lower left direction of this point is statistically counted to construct the gray-level co-occurrence matrix in the 135° direction; read the gray-level value of the pixel point (x, y) and the gray-level value of the adjacent pixel point (x - 1, y - 1) in the lower left direction of this point, and according to the same operation method above, traverse all pixel points of the entire gray-scale image of the rock wool board surface to obtain the gray-level co-occurrence matrix in the 135° direction; this direction can capture the texture change characteristics on the lower left direction of the rock wool board surface, which is very crucial for capturing the texture characteristics of another oblique angle on the rock wool board surface. Combined with other directions, it can comprehensively capture the texture changes caused by fiber orientation, uneven compaction degree and various defects on the rock wool board surface, providing multi-dimensional data support for accurately judging the quality of the rock wool board;

[0030] Calculate the contrast value: According to the gray-level co-occurrence matrices in the 0°, 45°, 90° and 135° directions, calculate the contrast value of each matrix respectively, all of which are obtained through the contrast calculation formula, as shown in Equation (5):

[0031]

[0032] where N g is the total number of gray levels in the gray-scale image of the rock wool board surface, P(i, j) is the element value at (i, j) in the gray-level co-occurrence matrix, and Contrast is the calculated contrast value; after calculating the contrast value of each matrix using Equation (5), we get Contrast 0° 、Contrast 45° 、Contrast 90° and Contrast 135°, and then the average contrast value is calculated as shown in Equation (6):

[0033]

[0034] Among them, Contrast norm is the calculated average contrast value, representing the total contrast value of the gray-scale image on the surface of the rock wool board. The larger the Contrast norm , the more obvious the defects on the surface of the rock wool board. The smaller the Contrast norm , the finer the defects on the surface of the rock wool board.

[0035] Preferably, the steps of constructing and training the rock wool board quality detection model based on the convolutional neural network and the support vector machine in step S40 include:

[0036] Dataset construction: Collect gray-scale image samples of normal rock wool boards without defects and gray-scale image samples of rock wool boards with various quality defects, and label the defects and feature information on the images. Divide them into a training set, a validation set, and a test set according to the ratio of 70%:15%:15%;

[0037] Model construction: The entire network model includes a convolutional neural network part and a support vector machine part. The output of the convolutional neural network is used as the input of the support vector machine; the convolutional neural network part includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer; during model training, the input of the input layer is the training set divided in the dataset construction step; the convolutional layer extracts multi-scale features through convolutional kernels of different sizes, uses 3×3 and 5×5 convolutional kernels to capture details and features respectively, and uses ReLU as the activation function; the pooling layer uses 2×2 maximum pooling operation to reduce the resolution of the feature map and reduce the model calculation amount, while retaining the key feature information; the fully connected layer integrates the features extracted by the convolutional layer and the pooling layer and inputs them into the support vector machine part; the support vector machine separates different types of defect samples in the high-dimensional space by constructing an optimal classification hyperplane and outputs, realizing the classification and recognition of different types of defects;

[0038] Model training and validation: After the model construction is completed, set the model parameters, use the divided training set as the input to train the model. During the training process, continuously adjust the parameters such as the weights and biases of the convolutional kernels in the convolutional layer of the convolutional neural network, and the Lagrange multipliers and biases in the support vector machine through backpropagation, so that the model learns the mapping relationship between the rock wool board image features and quality defects. After each round of training ends, use the divided validation set to validate the trained model;

[0039] Model evaluation: When overfitting occurs, stop the model training and validation operations, and use the divided test set for model evaluation;

[0040] Model optimization: Adjust the model parameters according to the model evaluation results and retrain until the optimal combination of model parameters is obtained, and determine the corresponding version of the rock wool board quality detection model.

[0041] Among them, in the construction of the data set, there are image samples of rock wool boards with various quality defects, and various quality defect types include:

[0042] Fiber exposure: In the fiber exposure area, the surface fibers show an irregular distribution state, which is manifested as uneven gray value distribution in the gray-scale image of the rock wool board, and the contrast value is higher than the surrounding area of the fiber exposure area;

[0043] Hole: The gray value in the hole area of the gray-scale image of the rock wool board is uniform and lower than the surrounding area of the hole area, and at the same time, the edge presents a circular or elliptical shape;

[0044] Lamination: The lamination is located in the side area of the rock wool board and presents a boundary line with different gray levels in the gray-scale image of the rock wool board;

[0045] Stain: The gray value in the stain area of the gray-scale image of the rock wool board is different from that of the surrounding area of the stain area, and the texture clarity is lower than that of the surrounding area of the stain area.

[0046] Preferably, in step S50, the number of defects refers to the number of various defects detected by the rock wool board quality detection model, and the severity of the defects is judged by the total contrast of the gray-scale image of the rock wool board surface. A defect number threshold T num and a total contrast threshold T contrast of the gray-scale image of the rock wool board surface are preset. When the number of detected defects is less than T num and the total contrast of the gray-scale image of the rock wool board surface calculated during the detection process is also less than T contrast , it is determined that the quality of the detected rock wool board is qualified.

[0047] In addition, to achieve the above object, the present invention also proposes a system for detecting the quality of rock wool boards, and the system for detecting the quality of rock wool boards includes:

[0048] Rock wool board surface image acquisition module: used to obtain the rock wool board surface image data by taking pictures of the rock wool board surface through an industrial camera;

[0049] Rock wool board surface image data processing module: used to perform data preprocessing on the obtained rock wool board surface image data, including grayscale processing and filtering denoising, to obtain the rock wool board surface grayscale image;

[0050] Rock wool board surface grayscale image texture feature extraction module: used to extract the texture features of the obtained rock wool board surface grayscale image to obtain the rock wool board surface grayscale image with the texture features extracted;

[0051] Rock wool board quality detection model and training module: used to collect grayscale image samples of normal rock wool boards without defects and grayscale image samples of rock wool boards with various quality defects, construct a rock wool board quality detection model based on convolutional neural network and support vector machine and train it;

[0052] Rock wool board quality detection module: after the model training is completed, input the grayscale image of the rock wool board surface with texture features obtained from the rock wool board surface grayscale image texture feature extraction module into the model for rock wool board quality detection and output the detection result;

[0053] The industrial camera used in the rock wool board surface image acquisition module has a 2k resolution. An annular light source is installed around the industrial camera, and the rock wool board is photographed from the front and side to obtain rock wool board surface image data at different angles;

[0054] The data preprocessing steps in the rock wool board surface image data processing module include grayscale processing and filtering denoising. Grayscale processing is used to simplify the data and display the texture features of the rock wool board surface; Filtering denoising is used to remove noise interference and retain the edge features and texture features of the rock wool board;

[0055] When outputting the detection result in the rock wool board quality detection module, the position where the defect exists and the information of the defect type are marked on the rock wool board surface image through visualization software and a detection report is generated; At the same time, according to the preset rock wool board quality standard, when the number of detected defects and the severity of the defects are within the range of the rock wool board quality standard, it is determined that the rock wool board quality is qualified; When the number of detected defects and the severity of the defects exceed the range of the rock wool board quality standard, it is determined that the rock wool board quality is unqualified and targeted improvement suggestions are generated.

[0056] In addition, to achieve the above object, the present invention also proposes a device for rock wool board quality detection, the device includes: a memory, a processor, and programs such as a rock wool board quality detection algorithm based on convolutional neural network and support vector machine stored on the memory and executable on the processor, and the programs such as the rock wool board quality detection algorithm based on convolutional neural network and support vector machine are steps for implementing a method for rock wool board quality detection as described above.

[0057] In addition, to achieve the above object, the present invention also provides a computer program product, the computer program product includes programs such as a rock wool board quality detection algorithm based on convolutional neural network and support vector machine, and when the programs such as the rock wool board quality detection algorithm based on convolutional neural network and support vector machine are executed by a processor, a method for rock wool board quality detection as described above is implemented.

[0058] The advantages and effects of the present invention are:

[0059] A method for detecting the quality of rock wool boards proposed by the present invention captures multiple features by combining image data processing with convolutional neural networks and support vector machines, accurately detects the quality of rock wool boards, and through automated image processing and artificial intelligence models for detection, without manual visual inspection and manual participation, greatly saving time and labor costs, shortening the detection time for the quality detection of rock wool boards, and improving the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0061] Figure 1 It is a flowchart of a method for detecting the quality of rock wool boards according to the present invention.

[0062] Figure 2 It is a schematic structural diagram of a system for detecting the quality of rock wool boards according to the present invention.

[0063] Figure 3 It is a schematic block diagram of the structure of an electronic device for detecting the quality of rock wool boards according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0064] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0065] The present invention provides a method for detecting the quality of rock wool boards, as Figure 1 shown, including the following steps:

[0066] Step S10: Take pictures from different directions and angles through multiple high-definition industrial cameras to ensure that the entire surface of the rock wool board is covered, and obtain the surface image data of the rock wool board.

[0067] Among them, the industrial cameras used in step S10 have a resolution of 2k, and the camera positions and shooting parameters are optimized according to the standard size of the rock wool board and the distribution characteristics of part defects. For example, for the delamination defects that are likely to occur on the side of the rock wool board, a low-angle oblique light camera is arranged on the side to highlight the layer characteristics; for the defects such as fiber exposure and holes commonly found on the top surface, a vertical top shooting camera combined with a ring light source is used to ensure uniform illumination and clearly capture the details.

[0068] Step S20: Perform data preprocessing on the obtained surface image data of the rock wool board, including grayscale processing and filtering denoising, to obtain the grayscale image of the rock wool board surface.

[0069] The data preprocessing steps in Step S20 include grayscale processing and filtering denoising. Grayscale processing is used to simplify the data and display the texture features of the rock wool board surface; filtering denoising is used to remove noise interference and retain the edge features and texture features of the rock wool board.

[0070] Specifically, the steps for performing data preprocessing on the obtained grayscale image of the rock wool board surface in Step S20 are as follows:

[0071] Grayscale processing: Using the weighted average method, for the obtained surface image data of the rock wool board, read the information of the red, green, and blue color channels of each pixel point in the image, and calculate the grayscale value of each pixel point, as shown in Equation (1):

[0072] Gray = 0.299 * R + 0.587 * G + 0.114 * B (1)

[0073] Where Gray is the grayscale value obtained through weighted average calculation, and this value is also the pixel value of the corresponding pixel point in the grayscale image. R is the color value of the red channel in the surface image data of the rock wool board, reflecting the brightness information of the pixel point in the red spectrum. G is the color value of the green channel in the surface image data of the rock wool board, reflecting the brightness information of the pixel point in the green spectrum. B is the color value of the blue channel in the surface image data of the rock wool board, reflecting the brightness information of the pixel point in the blue spectrum; Replace the three color channel values of the original pixel point with the calculated grayscale value to obtain a single-channel grayscale image of the rock wool board surface;

[0074] Filtering denoising: For each pixel point (x, y) in the obtained grayscale image of the rock wool board surface, determine a rectangular neighborhood with a size of 3×3, 5×5, etc. Then the pixel points in the rectangular neighborhood are (x1, y1), (x2, y2),..., (x n , y n ), and the corresponding grayscale values are Gray1, Gray2,..., Gray n , where n represents that there are n pixel points in the rectangular neighborhood, and calculate the spatial Gaussian weight ω s (x, y, x k , y k ), as shown in Equation (2):

[0075]

[0076] Among them, (x, y) is the current pixel point in the rectangular neighborhood, (x k , yk ) is the k-th pixel point within the rectangular neighborhood, where 0 < k ≤ n, and σ s is the spatial standard deviation, representing the attenuation rate of the spatial weight, used to measure the spatial distance relationship between the pixel points within the rectangular neighborhood and the current pixel point (x, y). The closer the distance, the greater the weight. At the same time, calculate the gray Gaussian weight ω r (g(x, y), g(x k , y k ), as shown in Equation (3):

[0077]

[0078] where g(x, y) is the gray value of the current pixel point within the rectangular neighborhood, and g(x k , y k ) is the gray value corresponding to the k-th pixel point within the rectangular neighborhood, and σ r is the gray standard deviation, which determines the rate of change of the gray weight, used to measure the similarity of the gray values between the pixel points within the rectangular neighborhood and the current pixel point (x, y). The closer the gray values, the greater the weight. After calculating the spatial standard deviation and the gray standard deviation, multiply the spatial standard deviation and the gray standard deviation to obtain the filtering weight ω(x, y, x k , y k ). Calculate the filtered gray value Gray’(x, y), as shown in Equation (4):

[0079]

[0080] Traverse each point in the gray image of the rock wool board surface, and replace the gray value of the pixel point in the original gray image of the rock wool board surface with Gray’(x, y) to obtain the gray image of the rock wool board surface after filtering and denoising.

[0081] Step S30: Extract the texture features of the obtained gray image of the rock wool board surface to obtain the gray image of the rock wool board surface with the texture features extracted.

[0082] Specifically, in step S30, when extracting the texture features of the obtained gray image of the rock wool board surface, the method of constructing a gray-level co-occurrence matrix is adopted, including:[[]]

[0083] Construct the gray-level co-occurrence matrix in the 0° direction: Constructing the gray-level co-occurrence matrix in the 0° direction means that when constructing the gray-level co-occurrence matrix with a certain pixel point (x, y) in the gray-scale image of the rock wool board surface as the center, the frequency of the gray-level value combination of the adjacent pixel point (x + 1, y) directly to the right of this point is statistically counted to construct the gray-level co-occurrence matrix in the 0° direction; The specific operation is to read the gray-level value of the pixel point (x, y), set it as i, and at the same time read the gray-level value of its adjacent pixel point (x + 1, y) to the right, set it as j. In a two-dimensional matrix with all initial values being 0, that is, the gray-level co-occurrence matrix to be constructed in the 0° direction, find the corresponding positions (the i-th row, j-th column and the j-th row, i-th column, the gray-level co-occurrence matrix has symmetry), add 1 to the element value at this position, and traverse all pixel points of the entire gray-scale image of the rock wool board surface in the order from left to right and from top to bottom to obtain the gray-level co-occurrence matrix in the 0° direction; This direction can capture the texture change characteristics of the rock wool board surface in the horizontal direction. For example, for some rock wool boards with relatively regular fiber arrangements, the distribution of gray-level value combinations in the 0° direction is relatively regular, while when there are defects such as horizontal cracks or holes, the distribution of gray-level value combinations will show abnormalities;

[0084] Construct the gray-level co-occurrence matrix in the 45° direction: Constructing the gray-level co-occurrence matrix in the 45° direction means that when constructing the gray-level co-occurrence matrix with a certain pixel point (x, y) in the gray-scale image of the rock wool board surface as the center, the frequency of the gray-level value combination of the adjacent pixel point (x + 1, y + 1) at a 45° angle to the upper right of this point is statistically counted to construct the gray-level co-occurrence matrix in the 45° direction; Read the gray-level value of the pixel point (x, y) and the gray-level value of the adjacent pixel point (x + 1, y + 1) at a 45° angle to the upper right of this point, and according to the same operation method as above, traverse all pixel points of the entire gray-scale image of the rock wool board surface to obtain the gray-level co-occurrence matrix in the 45° direction; This direction can capture the texture change characteristics of the rock wool board surface at a 45° angle to the upper right. For example, when there are oblique scratches on the rock wool board surface or there is a special distribution trend of fibers in this direction due to the production process, the gray-level co-occurrence matrix in the 45° direction can reflect the change of the gray-level value combination, thereby assisting in judging the defect situation on the rock wool board surface;

[0085] Construct the gray-level co-occurrence matrix in the 90° direction: The 90° direction means that when constructing the gray-level co-occurrence matrix with a certain pixel point (x, y) in the gray-scale image of the rock wool board surface as the center, the frequency of the gray-level value combination of the adjacent pixel point (x, y + 1) in the vertical direction of this point is counted to construct the gray-level co-occurrence matrix in the 90° direction; read the gray-level value of the pixel point (x, y) and the gray-level value of the adjacent pixel point (x, y + 1) in the vertical direction of this point, and traverse all pixel points of the entire gray-scale image of the rock wool board surface according to the same operation method as above to obtain the gray-level co-occurrence matrix in the 90° direction; this direction can capture the texture change characteristics on the vertical direction of the rock wool board surface. For example, when there are delamination defects on the rock wool board, there are obvious hierarchical changes in the vertical direction, and the gray-level co-occurrence matrix in the 90° direction will show a probability distribution of gray-level value combinations that is completely different from that of the normal area, which helps to accurately identify such defects;

[0086] Construct the gray-level co-occurrence matrix in the 135° direction: The 135° direction means that when constructing the gray-level co-occurrence matrix with a certain pixel point (x, y) in the gray-scale image of the rock wool board surface as the center, the frequency of the gray-level value combination of the adjacent pixel point (x - 1, y - 1) in the lower left direction of this point is counted to construct the gray-level co-occurrence matrix in the 135° direction; read the gray-level value of the pixel point (x, y) and the gray-level value of the adjacent pixel point (x - 1, y - 1) in the lower left direction of this point, and traverse all pixel points of the entire gray-scale image of the rock wool board surface according to the same operation method as above to obtain the gray-level co-occurrence matrix in the 135° direction; this direction can capture the texture change characteristics on the lower left direction of the rock wool board surface, which is very crucial for capturing the texture characteristics of another oblique angle on the rock wool board surface. Combined with other directions, it can comprehensively capture the texture changes on the rock wool board surface caused by fiber orientation, uneven compaction degree and various defects, providing multi-dimensional data support for accurately judging the quality of the rock wool board;

[0087] Calculate the contrast value: According to the gray-level co-occurrence matrices in the 0°, 45°, 90° and 135° directions, calculate the contrast value of each matrix respectively, all of which are obtained through the contrast calculation formula, as shown in Equation (5):

[0088]

[0089] where N g is the total number of gray levels in the gray-scale image of the rock wool board surface, P(i, j) is the element value at (i, j) in the gray-level co-occurrence matrix, and Contrast is the calculated contrast value; after calculating the contrast value of each matrix respectively using Equation (5), Contrast 0° 、Contrast 45° 、Contrast 90° and Contrast 135°, and then calculate the average contrast value as shown in Equation (6):

[0090]

[0091] where Contrast norm is the calculated average contrast value, representing the total contrast value of the gray-scale image of the rock wool board surface. The larger the Contrast norm , the more obvious the defects on the rock wool board surface. The smaller the Contrast norm , the finer the defects on the rock wool board surface.

[0092] Step S40: Collect gray-scale image samples of normal rock wool boards without defects and gray-scale image samples of rock wool boards with various quality defects, and construct and train a quality detection model for rock wool boards based on a convolutional neural network and a support vector machine.

[0093] Specifically, the steps of constructing and training a quality detection model for rock wool boards based on a convolutional neural network and a support vector machine in Step S40 include:

[0094] Dataset construction: Collect gray-scale image samples of normal rock wool boards without defects and gray-scale image samples of rock wool boards with various quality defects, label the defects and feature information on the images, and divide them into a training set, a validation set, and a test set according to the ratio of 70%:15%:15%;

[0095] Model construction: The entire network model includes a convolutional neural network part and a support vector machine part. The output of the convolutional neural network is used as the input of the support vector machine; the convolutional neural network part includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer; during model training, the input of the input layer is the training set divided in the dataset construction step; the convolutional layer extracts multi-scale features through convolutional kernels of different sizes, uses 3×3 and 5×5 convolutional kernels to capture details and features respectively, and uses ReLU as the activation function; the pooling layer uses 2×2 max pooling operation to reduce the resolution of the feature map and reduce the model calculation amount, while retaining key feature information; the fully connected layer integrates the features extracted by the convolutional layer and the pooling layer and inputs them into the support vector machine part; the support vector machine separates different types of defect samples in a high-dimensional space by constructing an optimal classification hyperplane and outputs, realizing the classification and recognition of different types of defects;

[0096] Model training and validation: After the model is constructed, set the model parameters. Use the divided training set as the input to train the model. During the training process, continuously adjust the parameters such as the weights and biases of the convolutional kernels in the convolutional layers of the convolutional neural network, as well as the Lagrange multipliers and biases in the support vector machine through backpropagation, so that the model learns the mapping relationship between the image features of the rock wool board and the quality defects. After each round of training, use the divided validation set to validate the trained model;

[0097] Model evaluation: When overfitting occurs, stop the model training and validation operations, and use the divided test set for model evaluation;

[0098] Model optimization: Adjust the model parameters according to the model evaluation results and retrain until the optimal combination of model parameters is obtained, and determine the corresponding version of the rock wool board quality detection model.

[0099] Among them, in the construction of the dataset, there are rock wool board image samples with various quality defects. The various quality defect types include:

[0100] Fiber exposure: In the fiber exposure area, the surface fibers show an irregular distribution state, which is manifested as uneven gray value distribution in the gray image of the rock wool board, and the contrast value is higher than the surrounding area of the fiber exposure area;

[0101] Hole: In the hole area, the gray value in the gray image of the rock wool board is uniform and lower than the surrounding area of the hole area, and the edge shows a circular or elliptical shape;

[0102] Lamination: The lamination is located in the side area of the rock wool board and presents a dividing line with different gray levels in the gray image of the rock wool board;

[0103] Stain: In the stain area, the gray value in the gray image of the rock wool board is different from the surrounding area of the stain area and the texture clarity is lower than the surrounding area of the stain area.

[0104] Step S50: After the model training is completed, input the gray image of the rock wool board surface with the extracted texture features obtained in step S30 into the model for rock wool board quality detection and output the detection result.

[0105] Among them, when outputting the detection result in step S50, mark the position where the defect exists and the information of the defect type on the rock wool board surface image through visualization software and generate a detection report; at the same time, according to the preset rock wool board quality standard, when the number of detected defects and the severity of the defects are within the range of the rock wool board quality standard, it is determined that the quality of the rock wool board is qualified; when the number of detected defects and the severity of the defects exceed the range of the rock wool board quality standard, it is determined that the quality of the rock wool board is unqualified and targeted improvement suggestions are generated.

[0106] Specifically, the number of defects in step S50 refers to the number of various defects detected by the rock wool board quality detection model. The severity of the defects is judged by the total contrast of the gray-scale image of the rock wool board surface. A defect quantity threshold T num and the total contrast threshold T contrast of the gray-scale image of the rock wool board surface are preset. When the detected number of defects is less than T num and the total contrast of the gray-scale image of the rock wool board surface calculated during the detection process is also less than T contrast , it is determined that the quality of the detected rock wool board is qualified, and the detection result and detection time are stored.

[0107] In addition, the present invention also proposes a rock wool board quality detection system. Please refer to Figure 2 . The described rock wool board quality detection system includes:

[0108] Rock wool board surface image acquisition module: used to obtain rock wool board surface image data by taking pictures of the rock wool board surface through an industrial camera;

[0109] Rock wool board surface image data processing module: used to perform data preprocessing on the obtained rock wool board surface image data, including grayscale processing and filtering denoising, to obtain the gray-scale image of the rock wool board surface;

[0110] Rock wool board surface gray-scale image texture feature extraction module: used to extract the texture features of the obtained gray-scale image of the rock wool board surface to obtain the gray-scale image of the rock wool board surface with texture features extracted;

[0111] Rock wool board quality detection model and training module: used to collect normal rock wool board gray-scale image samples without defects and rock wool board gray-scale image samples with various quality defects, construct a rock wool board quality detection model based on a convolutional neural network and a support vector machine, and train it;

[0112] Rock wool board quality detection module: used to input the gray-scale image of the rock wool board surface with texture features obtained in the rock wool board surface gray-scale image texture feature extraction module into the model for rock wool board quality detection after the model training is completed and output the detection result;

[0113] Among them, the industrial camera used in the rock wool board surface image acquisition module has a 2k resolution. An annular light source is installed around the industrial camera, and the rock wool board is photographed from the front and side to obtain rock wool board surface image data at different angles;

[0114] Among them, the data preprocessing steps in the rock wool board surface image data processing module include grayscale processing and filtering denoising. Grayscale processing is used to simplify the data and display the texture features of the rock wool board surface; filtering denoising is used to remove noise interference and retain the edge features and texture features of the rock wool board;

[0115] Among them, when the rock wool board quality detection module outputs the detection result, the visualization software marks the position where the defect exists and the information of the defect type on the surface image of the rock wool board and generates a detection report. At the same time, according to the preset rock wool board quality standard, when the number of detected defects and the severity of the defects are within the range of the rock wool board quality standard, it is determined that the quality of the rock wool board is qualified; when the number of detected defects and the severity of the defects exceed the range of the rock wool board quality standard, it is determined that the quality of the rock wool board is unqualified and targeted improvement suggestions are generated.

[0116] A rock wool board quality detection system provided by the present application adopts a method for detecting the quality of a rock wool board in the above-mentioned embodiment, which can solve the technical problems of low efficiency and low accuracy of the traditional rock wool board quality detection method. Compared with the prior art, the beneficial effects of a rock wool board quality detection system provided by the present application are the same as those of a method for detecting the quality of a rock wool board provided by the above-mentioned embodiment, and other technical features in the rock wool board quality detection system are the same as the features disclosed in the method of the above-mentioned embodiment, and will not be elaborated herein.

[0117] The present application provides a rock wool board quality detection device, and the rock wool board quality detection device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a method for detecting the quality of a rock wool board in the first embodiment above.

[0118] Reference is made below to Figure 3 , which shows a schematic structural diagram of a rock wool board quality detection device suitable for implementing the embodiments of the present application. A rock wool board quality detection device in the embodiments of the present application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions: tablet computers), PMPs (Portable Media Players), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The rock wool board quality detection device shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of the present application.

[0119] Figure 3A quality inspection device for rock wool boards shown can include a processing system 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage system 1003 into a random access memory (RAM: Random Access Memory) 1004. In the RAM 1004, various programs and data required for the operation of the quality inspection device for rock wool boards are also stored. The processing system 1001, the ROM 1002, and the RAM 1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input system 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output system 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage system 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication system 1009. The communication system 1009 can allow the quality inspection device for rock wool boards to communicate with other devices wirelessly or wiredly to exchange data. Although a quality inspection device for rock wool boards with various systems is shown in the figure, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems can be implemented or had alternatively.

[0120] In particular, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through the communication system, or installed from the storage system 1003, or installed from the ROM 1002. When the computer program is executed by the processing system 1001, the above-mentioned functions defined in the methods of the embodiments disclosed in the present application are executed.

[0121] A quality inspection device for rock wool boards provided by the present application adopts the method for quality inspection of rock wool boards in the above-mentioned embodiment, and can solve the technical problems of low efficiency and low accuracy of the traditional method for quality inspection of rock wool boards. Compared with the prior art, the beneficial effects of the quality inspection device for rock wool boards provided by the present application are the same as those of the method for quality inspection of rock wool boards provided by the above-mentioned embodiment, and other technical features in the quality inspection device for rock wool boards are the same as the features disclosed in the method of the previous embodiment, and will not be elaborated here.

[0122] The various parts disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0123] This application also provides a computer program product, including a computer program, and the steps of a method for quality inspection of rock wool boards as described above are implemented when the computer program is executed by a processor.

[0124] The computer program product provided by this application can solve the technical problems of low efficiency and low accuracy of traditional methods for quality inspection of rock wool boards. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as those of the method for quality inspection of rock wool boards provided in the above embodiments, and will not be elaborated here.

[0125] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations.

Claims

1. A method for quality inspection of rock wool boards, characterized in that, The method includes the following steps: Step S10: Obtain the surface image data of the rock wool board by taking pictures of the surface of the rock wool board with an industrial camera; Step S20: Perform data preprocessing on the obtained surface image data of the rock wool board, including grayscale processing and filtering denoising, to obtain the grayscale image of the surface of the rock wool board; Step S30: Extract the texture features of the grayscale image of the surface of the rock wool board obtained, to obtain the grayscale image of the surface of the rock wool board with the texture features extracted; Step S40: Collect the grayscale image samples of normal rock wool boards without defects and the grayscale image samples of rock wool boards with various quality defects, and construct and train a rock wool board quality detection model based on a convolutional neural network and a support vector machine; Step S50: After the model training is completed, input the grayscale image of the surface of the rock wool board with the texture features extracted in Step S30 into the model for rock wool board quality detection and output the detection result; The industrial camera used in Step S10 has a 2k resolution. An annular light source is installed around the industrial camera, and the rock wool board is photographed from the front and side to obtain the surface image data of the rock wool board at different angles; The data preprocessing steps in Step S20 include grayscale processing and filtering denoising. Grayscale processing is used to simplify the data and display the texture features of the surface of the rock wool board; Filtering denoising is used to remove noise interference and retain the texture features of the rock wool board; When outputting the detection result in Step S50, use visualization software to mark the position where the defect exists and the information of the defect type on the surface image of the rock wool board and generate a detection report; At the same time, according to the preset quality standard of the rock wool board, when the number and severity of the detected defects are within the range of the quality standard of the rock wool board, it is determined that the quality of the rock wool board is qualified; When the number and severity of the detected defects exceed the range of the quality standard of the rock wool board, it is determined that the quality of the rock wool board is unqualified and targeted improvement suggestions are generated.

2. The method for quality inspection of rock wool boards according to claim 1, wherein, The steps for performing data preprocessing on the obtained grayscale image of the surface of the rock wool board in Step S20 include: Grayscale processing: Adopt the weighted average method. For the obtained surface image data of the rock wool board, read the information of the red, green, and blue color channels of each pixel point in the image, and calculate the grayscale value of each pixel point, as shown in Equation (1): Gray = 0.299 * R + 0.587 * G + 0.114 * B (1) Where Gray is the grayscale value obtained after weighted average calculation, and this value is also the pixel value of the corresponding pixel point in the grayscale image. R is the color value of the red channel in the surface image data of the rock wool board, which reflects the brightness information of this pixel point in the red spectrum. G is the color value of the green channel in the surface image data of the rock wool board, which reflects the brightness information of this pixel point in the green spectrum. B is the color value of the blue channel in the surface image data of the rock wool board, which reflects the brightness information of this pixel point in the blue spectrum; Replace the three color channel values of the original pixel point with the calculated grayscale value to obtain a single-channel grayscale image of the surface of the rock wool board; Filter denoising: For each pixel point (x, y) in the obtained grayscale image of the rock wool board surface, a rectangular neighborhood is determined. The pixel points in the rectangular neighborhood are (x1, y1), (x2, y2),..., (x n , y n ), and the corresponding grayscale values are Gray1, Gray2,..., Gray n , where n represents that there are n pixel points in the rectangular neighborhood. Calculate the spatial Gaussian weight ω s (x, y, x k , y k ), as shown in Equation (2): where (x, y) is the current pixel point within the rectangular neighborhood, and (x k , y k ) is the k-th pixel point within the rectangular neighborhood, where 0 < k ≤ n, and σ s is the spatial standard deviation; simultaneously calculate the gray-scale Gaussian weight ω r (g(x, y), g(x k , y k )) as shown in Equation (3): Among them, g(x, y) is the gray value of the current pixel point in the rectangular neighborhood, and g(x k , y k ) is the gray value corresponding to the k-th pixel point in the rectangular neighborhood, and σ r is the standard deviation of gray. After calculating the spatial standard deviation and the standard deviation of gray, multiply the spatial standard deviation and the standard deviation of gray to obtain the filtering weight ω(x, y, x k , y k ). The filtered gray value Gray’(x, y) is obtained through calculation, as shown in Equation (4): Traverse each point in the grayscale image of the surface of the rock wool board, and replace the grayscale value of the pixel point in the original grayscale image of the surface of the rock wool board with Gray’(x,y) to obtain the grayscale image of the surface of the rock wool board after filtering denoising.

3. The method for quality inspection of rock wool boards according to claim 1, characterized in that, In step S30, feature extraction is performed on the texture features of the obtained grayscale image of the rock wool board. The method of constructing a gray-level co-occurrence matrix is adopted, including: Constructing a gray-level co-occurrence matrix in the 0° direction: In the 0° direction, it means that when constructing a gray-level co-occurrence matrix with a certain pixel point (x, y) in the grayscale image of the rock wool board surface as the center, the frequency of the gray value combination of the adjacent pixel point (x + 1, y) directly to the right of this point is statistically calculated to construct the gray-level co-occurrence matrix in the 0° direction; Constructing a gray-level co-occurrence matrix in the 45° direction: In the 45° direction, it means that when constructing a gray-level co-occurrence matrix with a certain pixel point (x, y) in the grayscale image of the rock wool board surface as the center, the frequency of the gray value combination of the adjacent pixel point (x + 1, y + 1) at a 45° angle to the upper right of this point is statistically calculated to construct the gray-level co-occurrence matrix in the 45° direction; Constructing a gray-level co-occurrence matrix in the 90° direction: In the 90° direction, it means that when constructing a gray-level co-occurrence matrix with a certain pixel point (x, y) in the grayscale image of the rock wool board surface as the center, the frequency of the gray value combination of the adjacent pixel point (x, y + 1) in the vertical direction of this point is statistically calculated to construct the gray-level co-occurrence matrix in the 90° direction; Constructing a gray-level co-occurrence matrix in the 135° direction: In the 135° direction, it means that when constructing a gray-level co-occurrence matrix with a certain pixel point (x, y) in the grayscale image of the rock wool board surface as the center, the frequency of the gray value combination of the adjacent pixel point (x - 1, y - 1) in the lower left direction of this point is statistically calculated to construct the gray-level co-occurrence matrix in the 135° direction; Calculating the contrast value: According to the gray-level co-occurrence matrices in the 0°, 45°, 90°, and 135° directions, the contrast value of each matrix is calculated respectively, all obtained through the contrast calculation formula, as shown in Equation (5); where N g is the total number of gray levels in the gray-scale image of the rock wool board surface, P(i,j) is the element value at (i,j) in the gray-level co-occurrence matrix, and Contrast is the calculated contrast value; after calculating the contrast value of each matrix using Equation (5), Contrast 0° , Contrast 45° , Contrast 90° and Contrast 135° are obtained, and then the average contrast value is calculated as shown in Equation (6): Among them, Contrast norm is the calculated average contrast value, representing the total contrast value of the grayscale image on the surface of the rock wool board.

4. A method for quality inspection of rock wool boards according to claim 1, characterized in that, The steps of constructing and training a rock wool board quality detection model based on a convolutional neural network and a support vector machine in step S40 include: Dataset construction: Collect grayscale image samples of normal rock wool boards without defects and grayscale image samples of rock wool boards with various quality defects, and label the defect and feature information on the images. They are divided into a training set, a validation set, and a test set according to the ratio of 70%:15%:15%; Model construction: The entire network model includes a convolutional neural network part and a support vector machine part. The output of the convolutional neural network is used as the input of the support vector machine; the convolutional neural network part includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer; during model training, the input of the input layer is the training set divided in the dataset construction step; the convolutional layer extracts multi-scale features through convolutional kernels of different sizes, uses 3×3 and 5×5 convolutional kernels to capture details and features respectively, and uses ReLU as the activation function; the pooling layer uses a 2×2 max pooling operation; the fully connected layer integrates the features extracted by the convolutional layer and the pooling layer and inputs them into the support vector machine part; the support vector machine constructs an optimal classification hyperplane to separate defect samples of different categories in the high-dimensional space and outputs; Model training and validation: After the model is constructed, set the model parameters, and use the divided training set as the input to train the model. During the training process, continuously adjust the weights and biases of the convolutional kernels in the convolutional layer of the convolutional neural network, as well as the Lagrange multipliers and biases in the support vector machine through backpropagation, so that the model learns the mapping relationship between the image features of the rock wool board and the quality defects. After each round of training, use the divided validation set to validate the trained model; Model evaluation: Stop the model training and validation operations when overfitting occurs, and use the divided test set for model evaluation; Model optimization: Adjust the model parameters according to the model evaluation results and retrain until the optimal combination of model parameters is obtained, and determine the corresponding version of the rock wool board quality detection model.

5. A method for quality inspection of rock wool boards according to claim 4, characterized in that In the construction of the dataset, there are rock wool board image samples with various quality defects, and various quality defect types include: Fiber exposure: In the fiber exposure area, the surface fibers show an irregular distribution state, which is manifested as uneven gray value distribution in the gray-scale image of the rock wool board, and the contrast value is higher than the surrounding area of the fiber exposure area; Hole: The gray value in the hole area of the gray-scale image of the rock wool board is uniform and lower than the surrounding area of the hole area, and the edge shows a circular or elliptical shape at the same time; Lamination: The lamination is located in the side area of the rock wool board and presents a boundary line with different gray levels in the gray-scale image of the rock wool board; Stain: The gray value in the stain area of the gray-scale image of the rock wool board is different from the surrounding area of the stain area, and the texture clarity is lower than the surrounding area of the stain area.

6. The method for quality inspection of rock wool boards according to claim 1, characterized in that, The number of defects in step S50 refers to the number of various defects detected by the rock wool board quality detection model. The severity of the defects is judged by the total contrast of the grayscale image of the rock wool board surface. A defect quantity threshold T is preset in advance. num and the total contrast threshold T of the grayscale image of the rock wool board surface contrast , when the detected number of defects is less than T num and the total contrast of the grayscale image of the rock wool board surface calculated during the detection process is also less than T contrast , it is determined that the quality of the detected rock wool board is qualified.

7. A quality inspection system for rock wool boards, characterized in that, The described rock wool board quality detection system includes: Rock wool board surface image acquisition module: used to obtain rock wool board surface image data by taking pictures of the rock wool board surface with an industrial camera; Rock wool board surface image data processing module: used to perform data preprocessing on the obtained rock wool board surface image data, including grayscale processing and filtering denoising, to obtain the rock wool board surface grayscale image; Rock wool board surface grayscale image texture feature extraction module: used to extract the texture features of the obtained rock wool board surface grayscale image to obtain the rock wool board surface grayscale image with texture features extracted; Rock wool board quality detection model and training module: used to collect normal rock wool board gray-scale image samples without defects and rock wool board gray-scale image samples with various quality defects, construct a rock wool board quality detection model based on a convolutional neural network and a support vector machine, and train it; Rock wool board quality detection module: used to input the rock wool board surface grayscale image with texture features obtained in the rock wool board surface grayscale image texture feature extraction module into the model for rock wool board quality detection and output the detection result after the model training is completed; The industrial camera used in the rock wool board surface image acquisition module has a 2k resolution. An annular light source is installed around the industrial camera, and the rock wool board is photographed from the front and side to obtain rock wool board surface image data at different angles; The data preprocessing steps in the rock wool board surface image data processing module include grayscale processing and filtering denoising. Grayscale processing is used to simplify the data and display the texture features of the rock wool board surface; Filtering denoising is used to remove noise interference and retain the edge features and texture features of the rock wool board; When the quality inspection module of the rock wool board outputs the inspection results, the visualization software marks the location of the defect and the information of the defect type on the surface image of the rock wool board and generates an inspection report; at the same time, according to the preset quality standard of the rock wool board, when the number and severity of the detected defects are within the quality standard range of the rock wool board, it is determined that the quality of the rock wool board is qualified; when the number and severity of the detected defects exceed the quality standard range of the rock wool board, it is determined that the quality of the rock wool board is unqualified and targeted improvement suggestions are generated.

8. A quality inspection device for rock wool boards, characterized in that, The described quality inspection equipment for rock wool boards includes: A memory, a processor, and a quality inspection program for rock wool boards stored on the memory and executable on the processor. When the quality inspection program for rock wool boards is executed by the processor, it implements the quality inspection method for rock wool boards described in any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes a quality inspection program for rock wool boards. When the quality inspection program for rock wool boards is executed by the processor, it implements the quality inspection method for rock wool boards described in any one of claims 1 to 6.

Images