Intelligent detection method for surface cracks of laminated slab

Abstract

The invention relates to the technical field of image processing, in particular to a laminated slab surface crack intelligent detection method, which comprises the following steps: acquiring a surface image of a current laminated slab; clustering all the pixel points based on the texture roughness of each pixel point to obtain a plurality of image areas; calculating a first defect probability of each image area; obtaining a corresponding second defect probability based on the obtained sound wave signal of the surface of the laminated slab corresponding to each image area; fusing the first defect probability and the second defect probability corresponding to each image area by using the gray weight and the sound wave weight to obtain a defect degree; if the defect degree of at least one image area is greater than or equal to a threshold value, determining that the quality of the current laminated slab is unqualified; and if the defect degree of each image area is smaller than a threshold value, continuously performing quality judgment on the current laminated slab according to a set condition. According to the scheme, the surface crack of the laminated slab can be accurately detected.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing. More specifically, the present invention relates to an intelligent detection method for surface cracks of laminated plates. Background Art

[0002] A laminated plate is a precast concrete component that is widely used in modern construction projects. The application of laminated plates in buildings is directly related to the overall stability and safety of the building.

[0003] To ensure the safety and durability of building structures, it is crucial to detect laminated plates. Through detection, defects and problems that may occur during the production, transportation, and installation of laminated plates can be discovered in a timely manner, so as to take effective measures for treatment and avoid potential safety hazards. During factory production or construction, for the detection of laminated plates, the following common laminated plate detection methods are available: (1) Appearance inspection: The surface quality of the laminated plate, including obvious defects such as cracks, bubbles, and spalling, is detected by visual inspection or image processing technology. The disadvantage of visual inspection is that the manual detection efficiency is low. For large-area laminated plates, the detection speed is slow, consuming a large amount of manpower and time. At the same time, for some fine cracks, it is difficult for humans to detect them, which is prone to missed detection and affects the building quality. Although image detection technology can detect fine cracks and has high detection efficiency, it cannot detect crack defects related to the internal structure hidden in the laminated plate.

[0004] (2) Ultrasonic detection: Ultrasonic detection utilizes the propagation characteristics of sound waves in concrete to judge internal defects such as holes and cracks by detecting the reflection and transmission of sound waves. It is applicable to the detection of internal cracks in laminated plates. During detection, the propagation effect is easily affected by the flatness, cleanliness, and coupling agent of the concrete surface. At the same time, for small cracks or shallow cracks, the change of ultrasonic signals may not be obvious; therefore, its detection results may not fully and accurately reflect the true situation of the cracks.

[0005] Therefore, there is an urgent need for a detection method that can accurately detect cracks in laminated plates, effectively identify quality problems of laminated plates, and ensure the safety and reliability of building structures. Summary of the Invention

[0006] The purpose of the present invention is to propose an intelligent detection method for surface cracks of laminated plates to solve the problem of inaccurate crack detection results of laminated plates in the prior art; for this purpose, the present invention provides a solution in the following aspect.

[0007] An intelligent detection method for surface cracks of laminated plates provided by the present invention includes: Obtain the surface image of the current laminated plate; Obtain the texture roughness of each pixel point in the surface image; Cluster all pixel points based on their respective texture roughnesses to obtain multiple image regions; Input each image region into a crack recognition network to obtain a first defect probability; Use the ultrasonic method to obtain the acoustic wave signals on the surface of the laminated board corresponding to each image region, so as to obtain a second defect probability of the surface of the laminated board corresponding to each image region; Perform weighted summation on the first defect probability and the second defect probability corresponding to each image region according to the gray weight and the acoustic wave weight to obtain the defect degree; where the sum of the gray weight and the acoustic wave weight is 1, and the acoustic wave weight is positively correlated with the mean value of the texture roughness of all pixel points in each image region and the signal-to-noise ratio of the acoustic wave signal, and the texture roughness characterizes the texture complexity of each image region; If the defect degree of at least one image region is greater than or equal to the threshold, the quality of the current laminated board is unqualified; if the defect degree of each image region is less than the threshold, continue to judge the quality of the current laminated board according to the set conditions.

[0008] In the above solution, the image regions with similar textures are initially divided by texture roughness clustering, and the first defect probability of the image regions is obtained. Then, in combination with the acoustic wave signals obtained by ultrasonic detection, the second defect probability is determined. Finally, the two defect judgment results are weighted and fused to obtain the accurate defect degree of the cracks in the laminated board, and the cracks in the laminated board can be accurately detected.

[0009] Optionally, the acoustic wave weight is the product of the normalized value of the mean value of the texture roughness and the normalized value of the signal-to-noise ratio of the acoustic wave signal.

[0010] In the above solution, the defect results obtained by different detection methods are fused using the gray weight and the acoustic wave weight, and the weight is dynamically adjusted according to the texture roughness, so that the detection result is more in line with the actual crack situation, and the accuracy of the detection is further improved.

[0011] Optionally, the texture roughness is the gradient entropy of each pixel point, and the gradient entropy is the information entropy of the gradient between each pixel point and its neighboring pixel points within a set window.

[0012] In the above solution, by obtaining the gradient entropy of the pixel points, the degree of chaos of the pixel points can be characterized.

[0013] Optionally, the second defect probability is negatively correlated with the similarity of the corresponding image region; the similarity is the similarity degree between the calculated wavelet coefficient sequence of the image region and the standard wavelet coefficient sequence of the standard laminated board, and the wavelet coefficient sequence is obtained by performing wavelet analysis on the acoustic wave signals corresponding to each image region.

[0014] By comparing the surface of the laminated plate corresponding to the image region with the standard laminated plate, the relative defect condition of the local surface of the laminated plate corresponding to the image region can be determined.

[0015] Optionally, the second defect probability is the normalized value of the depth obtained from the acoustic wave signal.

[0016] Optionally, the crack recognition network is a convolutional neural network model, and its training process is as follows: Obtain a training set, which includes a historical laminated plate image set and labels; Input the training set into the network prediction model for training, calculate the loss value using a loss function, and use the gradient descent algorithm to adjust the parameters of the network prediction model until the loss value between the output prediction value and the label is less than the threshold or the number of training times reaches the set number of times, then stop training and obtain the trained network prediction model.

[0017] Optionally, continuing to judge the quality of the current laminated plate according to the set conditions includes: When the set conditions are met, the quality of the current laminated plate is unqualified; when the set conditions are not met, the laminated plate needs to be repaired; The set conditions are that there is at least one crack with a length greater than or equal to the length threshold and / or there is at least one crack with an area greater than or equal to the area length threshold.

[0018] The above solution can further perform refined detection on the current laminated plate, improving the accuracy of crack detection of the current laminated plate.

[0019] Optionally, the length and area of the crack include: Use an edge detection algorithm to obtain multiple edge information in the surface image, perform morphological operations on the multiple edge information to obtain the crack edge contour; and use the contour tracking algorithm to obtain the contour point sequence of the crack edge contour, calculate the distance between adjacent two contour points in the contour point sequence, and take the sum of all distances as the length of the crack; Obtain the minimum bounding rectangle of the crack edge contour, and take the area of the minimum bounding rectangle as the area of the crack.

[0020] Optionally, the clustering method uses the K-means clustering algorithm, and the number of clusters is determined by the elbow method.

[0021] Optionally, it further includes the steps of performing Gaussian filtering processing and grayscale processing on the surface image.

[0022] The beneficial effects of the present invention are: The solution of the present invention comprehensively applies two technologies, namely image texture analysis and ultrasonic detection, to judge the crack situation from two aspects of surface texture features and internal structure features, which can more comprehensively and accurately evaluate the defect degree of cracks, avoid the limitations of a single detection method, and reduce the situations of misjudgment and missed judgment. Compared with relying only on image analysis, which may ignore internal hidden cracks, or relying only on ultrasonic detection, which is insensitive to surface micro-cracks, this method can accurately identify various crack situations. Brief Description of the Drawings

[0023] Figure 1 Schematically shows the step flowchart of an intelligent detection method for surface cracks of a composite slab in this embodiment. Detailed Implementation Manner

[0024] The scenario of the present invention can be the detection of composite slabs at the factory or the detection of composite slabs at the construction site.

[0025] Taking the detection of composite slabs at the factory as an example, an intelligent detection method for surface cracks of a composite slab in this embodiment is introduced. Specifically, as Figure 1 shown, it includes the following steps: Step S1, obtain the surface image of the current composite slab.

[0026] In this embodiment, the surface image of the composite slab is obtained through an image acquisition device. Among them, the image acquisition device can select a high-resolution industrial camera, and its resolution should be determined according to the size of the composite slab and the minimum crack width expected to be detected. For example, for composite slabs of common sizes and expecting to detect cracks with a width of 0.1 mm, the camera resolution can be selected to be 5 million pixels or more to ensure that the details of the surface of the composite slab can be clearly presented in the collected image.

[0027] At the same time, to ensure the stability of image acquisition, the camera should be equipped with a stable bracket, and the lens should be selected with a smaller distortion model to reduce the impact of image deformation on subsequent analysis.

[0028] It should be noted that the lighting conditions of the acquisition environment have a great impact on the image quality. It is recommended to select an environment with uniform light and moderate brightness as much as possible to ensure that there are no shadows, reflections, etc. on the surface of the composite slab.

[0029] In this embodiment, the surface image is also subjected to Gaussian filtering processing and grayscale processing to obtain the preprocessed surface image.

[0030] Step S2, divide the surface image to obtain multiple image regions.

[0031] In one embodiment, the process of dividing the surface image is as follows: First, obtain the texture roughness of each pixel point in the surface image.

[0032] The texture roughness in this embodiment is the gradient entropy of each pixel point; where the gradient entropy is the information entropy of the gradient between any pixel point and its neighboring pixel points within a set window. The information entropy of the gradient is the information entropy of the gradient direction or the information entropy of the gradient value; it can also be the mean of the information entropy of the gradient direction and the information entropy of the gradient value.

[0033] Among them, the information entropy reflects the uncertainty of the information in the image. The more uniform the texture, the smaller the information entropy. The information entropy more comprehensively reflects the degree of texture uniformity from the perspective of information theory. Since the acquisition of the information entropy is a prior art, it will not be elaborated here.

[0034] The above set window can be 5×5 or 7×7; of course, it can also be determined according to the actual situation.

[0035] As other embodiments, a method combining local binary pattern (LBP) with statistical analysis can also be adopted. LBP is an operator used to describe the local texture features of an image. By comparing the gray values of the central pixel and its neighboring pixels, a binary code is generated to reflect the local texture pattern. After calculating the LBP features for each image region, statistical quantities such as the variance of the LBP histogram can be statistically analyzed to characterize the texture roughness.

[0036] Secondly, all pixel points are clustered based on the texture roughness of each pixel point to obtain multiple image regions.

[0037] In this embodiment, based on the texture roughness of each pixel point, all pixel points are clustered and analyzed to obtain multiple clusters, and each cluster is an image region. Among them, the mean value of all texture roughnesses within each image region is used as the regional texture roughness of the corresponding image region.

[0038] In one embodiment, the clustering algorithm can adopt the K-Means clustering algorithm; it should be noted that when clustering, the number of clusters is determined by the elbow method.

[0039] Since the K-Means clustering algorithm and the elbow method are prior arts, they will not be elaborated here.

[0040] In another embodiment, the clustering algorithm can also adopt the hierarchical clustering algorithm. The hierarchical clustering algorithm does not require specifying the number of clusters in advance and can determine the appropriate number of clusters by observing the dendrogram of the clustering result. For complex situations such as the surface image of a laminated plate, it can perform clustering analysis more flexibly.

[0041] The reason for clustering based on the texture roughness of each pixel is that cracks will cause sudden changes in the surface texture, so the texture roughness of cracks is relatively large; while the texture of normal areas is relatively smooth and the texture roughness is small. Therefore, through clustering, pixel points with similar texture roughness can be grouped into one category, and the image can be segmented into multiple regions with similar texture or structural characteristics, that is, the texture complex regions are divided into one category, which better meets the actual detection requirements.

[0042] The purpose of dividing the surface image in the above embodiments is to facilitate more detailed analysis of the image. Each similar small region is relatively easier to process and analyze its features compared to the overall image, which helps to improve the accuracy and efficiency of detection.

[0043] Step S3, calculate the defect degree of each image region.

[0044] Specifically, the process of obtaining the defect degree of each image region includes steps S31 - S33, specifically as follows: Step S31, calculate the first defect probability of each image region.

[0045] In this embodiment, each image region is input into the crack recognition network to obtain the first defect probability.

[0046] The crack recognition network therein is a convolutional neural network model, such as a CNN network.

[0047] Taking the CNN network as an example, its training process is as follows: Obtain a training set, which includes a historical set of laminated plate images and labels; the labels are marked for the cracks in the image set manually. For example, the labels can be divided into 3 categories: normal, minor defects (surface cracks), and severe defects (deep cracks).

[0048] Input the training set into the CNN network for training, and use a loss function to calculate the loss value. Use the gradient descent algorithm to adjust the parameters of the CNN network until the loss value between the output prediction value and the label is less than the threshold or the number of training times reaches the set number, then stop training and obtain the trained CNN network. Among them, the loss function is the cross - entropy loss.

[0049] After obtaining the trained CNN network, input each image region into the trained CNN network to obtain the first defect probability of the image region.

[0050] Step S32, use the ultrasonic method to obtain the acoustic wave signal on the surface of the laminated plate corresponding to each image region, and determine the second defect probability on the surface of the laminated plate of each image region based on the acoustic wave signal.

[0051] In this embodiment, the acoustic wave signals on the surface of the laminated slab corresponding to the image region are obtained by the ultrasonic method. Among them, the working principle of the ultrasonic method is as follows: when ultrasonic waves propagate inside the laminated slab, reflections, refractions, etc. will occur when encountering different media or structural changes, thereby obtaining acoustic wave signals. According to the characteristics of these signals, such as waveform, amplitude, etc., it can be analyzed whether there are defects inside the surface of the laminated slab.

[0052] Among them, for the ultrasonic device, an ultrasonic probe with a suitable frequency is selected, and the selection of the frequency depends on the material of the laminated slab and the expected depth of the defect to be detected. For example, for a common concrete laminated slab, if defects with a relatively shallow depth (such as 1 - 5 cm) are to be detected, a high-frequency probe with a frequency of 5 - 10 MHz can be selected to obtain a higher resolution; if defects with a relatively deep depth (such as 5 - 15 cm) are to be detected, a low-frequency probe with a frequency of 1 - 5 MHz is selected to ensure that the ultrasonic waves have sufficient penetration ability.

[0053] In one embodiment, a plurality of fixed measurement points can be set on the surface of the laminated slab corresponding to the image region, the acoustic wave signals of each fixed measurement point are measured, and finally the mean value of the acoustic wave signals of all fixed measurement points is used as the acoustic wave signal of the image region.

[0054] Since the sizes of different image regions may be different, the number of fixed measurement points is also different. For example, when the image region is larger, the number of fixed measurement points can be more; when the image region is smaller, the number of fixed measurement points can be less.

[0055] In another embodiment, the surface of the laminated slab corresponding to the image region can be directly scanned to obtain acoustic wave signals.

[0056] After obtaining the acoustic wave signals on the surface of the laminated slab corresponding to the image region, the corresponding second defect probability is calculated.

[0057] In one embodiment, in this embodiment, the time-frequency analysis method is adopted for each acoustic wave signal, and the analysis result is compared with the time-frequency result of the standard laminated slab to determine the second defect probability.

[0058] Among them, the time-frequency analysis method can be the short-time Fourier transform (STFT) or wavelet transform.

[0059] Specifically, taking the wavelet transform as an example, the second defect probability of each image region is calculated. Specifically, the wavelet transform is used to perform wavelet analysis on each acoustic wave signal and the acoustic wave signal of the standard laminated slab respectively to obtain the corresponding wavelet coefficient sequence and the standard wavelet coefficient sequence; the similarity degree between the corresponding wavelet coefficient sequence and the standard wavelet coefficient sequence is calculated, and the second defect probability of the image region is determined according to the similarity degree, where the second defect probability is negatively correlated with the similarity degree.

[0060] Among them, the greater the degree of similarity, the closer the local surface of the laminated plate in the image area corresponding to the wavelet coefficient sequence is to the standard laminated plate, the more likely the local surface of the laminated plate is a normal area, and the smaller the second defect probability; on the contrary, the smaller the degree of similarity, the more likely the local surface of the laminated plate is a crack area at this time, and the greater the second defect probability.

[0061] The above standard laminated plate is a normal and crack-free laminated plate, and this laminated plate belongs to the same type as the current laminated plate to avoid errors when calculating the degree of similarity. It should be noted that the size of the standard laminated plate can be the same as or different from the size of the local surface of the laminated plate corresponding to the image area. When the sizes are different, the degree of similarity can be obtained by calculating the shortest distance through the DTW algorithm. For example, the reciprocal of the shortest distance can be used as the degree of similarity.

[0062] Specifically, the second defect probability is: ; where is the degree of similarity between the wavelet coefficient sequence of the i-th image area and the standard wavelet coefficient sequence.

[0063] In the above embodiment, by analyzing the signal in the time and frequency domains, it helps to more clearly identify the characteristic information in the signal, so as to accurately determine the second defect probability on the surface of the laminated plate.

[0064] In another embodiment, considering that the crack depth is also one of the indicators for judging cracks, that is, the greater the depth of the crack, the more serious the crack on the local surface of the laminated plate, and the greater the second defect probability; on the contrary, the smaller the second defect probability; therefore, the second defect probability can also obtain the crack depth according to the acoustic wave signal; determine the second defect probability based on the depth, where the second defect probability is positively correlated with the depth. Specifically, the second defect probability can be the ratio of the depth to the thickness of the laminated plate.

[0065] Since obtaining the crack depth through the acoustic wave signal is a prior art, it will not be elaborated here.

[0066] In this embodiment, through the method of acoustic wave detection, some defects related to cracks hidden under the surface can be detected, improving the comprehensiveness of detection.

[0067] Step S33, calculate the acoustic wave weight and gray weight of each image area; fuse the first defect probability and the second defect probability according to the gray weight and acoustic wave weight of each image area to obtain the defect degree of the crack in the corresponding image area.

[0068] Among them, the echo received by the acoustic wave is related to the crack on the surface of the laminated slab. When there is an actual physical crack on the surface of the laminated slab, the texture roughness of the image at this time is rougher than that of a normal image. Therefore, when the roughness is greater, the acoustic wave weight should be greater. At the same time, considering the signal-to-noise ratio of the acoustic wave signal when obtaining the second defect probability, in this embodiment, the texture roughness and the signal-to-noise ratio are combined to obtain the acoustic wave weight. Specifically, the acoustic wave weight is positively correlated with the texture roughness and the signal-to-noise ratio.

[0069] In one embodiment, the acoustic wave weight is the product of the normalized value of the mean texture roughness and the normalized value of the signal-to-noise ratio of the acoustic wave signal.

[0070] In another embodiment, the acoustic wave weight is the mean of the normalized value of the mean texture roughness and the normalized value of the signal-to-noise ratio of the acoustic wave signal.

[0071] The above normalization uses the maximum-minimum normalization method. Among them, the sum of the gray weight and the acoustic wave weight is 1.

[0072] Among them, the defect degree is: ; is the gray weight of the i-th image region, is the acoustic wave weight of the i-th image region, is the first defect probability of the i-th image region, is the second defect probability of the i-th image region.

[0073] The above fusion method comprehensively considers the image texture features and the acoustic wave detection results, and can more accurately evaluate the severity of the crack.

[0074] Step S4, if the defect degree of at least one image region is greater than or equal to the threshold, the quality of the current laminated slab is unqualified; if the defect degree of each image region is less than the threshold, continue to judge the quality of the current laminated slab according to the set conditions.

[0075] In this embodiment, when the defect degree of at least one image region is greater than or equal to the threshold, it is considered that the defect of the current laminated slab is relatively serious, that is, the quality is unqualified.

[0076] When the defect degree of each image is less than the threshold, considering an image region corresponding to the defect degree, it may not be a complete crack. Therefore, it is also necessary to analyze the complete crack to finally determine the defect situation of the current laminated slab.

[0077] Specifically, when analyzing the complete crack, set the set conditions. When the set conditions are met, the quality of the current laminated slab is unqualified; when the set conditions are not met, the laminated slab needs to be repaired.

[0078] Among them, the set condition is that there is at least one crack with a length greater than or equal to the length threshold and / or there is at least one crack with an area greater than or equal to the area threshold.

[0079] Among them, when there are cracks in the current laminated slab, the process of obtaining the length and area of the cracks is as follows: Use an edge detection algorithm (such as the canny algorithm) to obtain multiple edge information in the surface image, perform morphological operations on the multiple edge information to obtain the crack edge contour; and use a contour tracking algorithm (such as the Moore-NeighhorTracing algorithm) to obtain the contour point sequence of the crack edge contour, and calculate the distance between two adjacent contour points in the contour point sequence, and take the sum of all distances as the length of the crack; obtain the minimum bounding rectangle of the crack edge contour, and take the area of the minimum bounding rectangle as the area of the crack.

[0080] The above distance is the Euclidean distance between two contour points.

[0081] The above length threshold can be obtained by obtaining the lengths of the cracks belonging to unqualified quality in the images of multiple historical laminated slabs, and taking the average value of all lengths as the length threshold; of course, it can also be taken according to experience; the above area threshold can be obtained by obtaining the areas of the cracks belonging to unqualified quality in the images of multiple historical laminated slabs, and taking the average value of all areas as the area threshold.

[0082] The above length threshold can also be determined according to the usage scenario of the laminated slab. For example, for laminated slabs used in key parts of high-rise buildings, the requirements for crack length are more stringent, and a lower threshold can be set; while for some secondary parts or application scenarios with lower crack sensitivity, the threshold for crack length can be appropriately increased.

[0083] The purpose of obtaining the above area is to consider the distribution of cracks. For example, whether the cracks are concentrated in a certain area (such as radioactive cracks, the crack length is limited, but in a radial shape). At this time, although the cracks are concentrated in a small area, they may also have a greater impact on the performance of the laminated slab. Therefore, the area of the cracks needs to be considered. At the same time, the crack length threshold in the qualified standard can also be correspondingly reduced.

[0084] Furthermore, based on image analysis to calculate the crack length, the defect position information obtained by ultrasonic testing can also be combined to calculate the crack length more accurately. For example, by spatially positioning the defect positions detected by ultrasonic testing, matching and fusing them with the crack positions obtained by image analysis, and using 3D reconstruction technology to construct a crack model on the surface of the laminated slab, so as to accurately measure the actual length of the crack in 3D space. This method can avoid the calculation error of the crack length caused by factors such as the image perspective.

[0085] The entire detection process of the solution of the present invention, from image acquisition, region division, feature calculation to final defect judgment, realizes automated operation, greatly reducing manual intervention. This not only improves the detection efficiency and enables rapid processing of the detection tasks of a large number of laminated plates, but also ensures the stability and reliability of the detection results. In large-scale building component production plants, rapid and accurate automated detection of laminated plates on the production line can be achieved.

[0086] Meanwhile, through a multi-level judgment process, clear judgment results and corresponding treatment suggestions are given for the quality status of the laminated plates. That is, it can not only directly judge whether the laminated plates are qualified, but also, for the case where there are cracks but the degree is relatively light, guide targeted repair, providing strong support for quality control during the building construction process and helping to improve the overall quality of the building project.

[0087] In the description of this specification, "a plurality of" means at least two, such as two, three or more, etc., unless otherwise specifically and clearly defined.

[0088] Although this specification has shown and described multiple embodiments of the present invention, it is obvious to those skilled in the art that such embodiments are provided only by way of example. Those skilled in the art will think of many changes, alterations and alternative ways without departing from the spirit and idea of the present invention. It should be understood that various alternative solutions to the embodiments of the present invention described herein can be adopted in the process of practicing the present invention.

Claims

1. An intelligent method for detecting surface cracks of a composite plate, characterized in that: include: Get the surface image of the current laminated plate; Clustering all pixels based on the texture roughness of each pixel to obtain multiple image regions; Input each image region into a crack recognition network to obtain a first defect probability; Acquiring the acoustic wave signal of the surface of the laminated plate corresponding to each image area by ultrasonic method, so as to obtain the second defect probability of the surface of the laminated plate corresponding to each image area; The first defect probability and the second defect probability corresponding to each image area are weighted and summed according to the grayscale weight and the acoustic wave weight to obtain the defect degree; wherein the sum of the grayscale weight and the acoustic wave weight is 1, the acoustic wave weight is positively correlated with the mean value of the texture roughness of all pixels in each image area and the signal-to-noise ratio of the acoustic wave signal, and the texture roughness represents the texture complexity of each image area; If the defect degree of at least one image area is greater than or equal to the threshold, the quality of the current superimposed board is unqualified; if the defect degree of each image area is less than the threshold, the quality judgment of the current superimposed board continues according to the set conditions.

2. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: The acoustic wave weight is the product of a normalized value of the mean value of the texture roughness and a normalized value of the signal-to-noise ratio of the acoustic wave signal.

3. The method for intelligent detection of surface cracks of a laminated plate according to claim 2, characterized in that: The texture roughness is the gradient entropy of each pixel point, and the gradient entropy is the information entropy of the gradient between each pixel point and its neighboring pixel points in a set window.

4. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: The second defect probability is negatively correlated with the similarity of the corresponding image area; the similarity is the degree of similarity between the calculated wavelet coefficient sequence of the image area and the standard wavelet coefficient sequence of the standard laminated plate, and the wavelet coefficient sequence is obtained by performing wavelet analysis on the acoustic wave signal corresponding to each image area.

5. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: The second defect probability is the ratio of the depth obtained by the acoustic wave signal to the thickness of the laminated plate.

6. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: The crack recognition network is a convolutional neural network model, and its training process is as follows: Obtaining a training set, wherein the training set includes a historical superimposed plate image set and labels; The training set is input into the convolutional neural network model for training, and the loss function is used to calculate the loss value. The parameters of the convolutional neural network model are adjusted using the gradient descent algorithm until the loss value between the output prediction value and the label is less than the threshold or the number of training times reaches the set number. The training is stopped and a trained convolutional neural network model is obtained.

7. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: The step of continuing to judge the quality of the current laminated plate according to the set conditions includes: When the set conditions are met, the quality of the current composite board is unqualified; when the set conditions are not met, the composite board needs to be repaired; The set condition is that the length of at least one crack is greater than or equal to a length threshold and / or the area of ​​at least one crack is greater than or equal to an area length threshold.

8. The method for intelligent detection of surface cracks of a laminated plate according to claim 7, characterized in that: The length of the crack and the area of ​​the crack include: An edge detection algorithm is used to obtain multiple edge information in the surface image, and morphological operations are performed on the multiple edge information to obtain the edge contour of the crack; and a contour tracking algorithm is used to obtain a contour point sequence of the edge contour of the crack, and the distance between two adjacent contour points in the contour point sequence is calculated, and the sum of all distances is taken as the length of the crack; The minimum bounding rectangle of the crack edge contour is obtained, and the area of ​​the minimum bounding rectangle is taken as the area of ​​the crack.

9. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: The clustering method adopts K-means clustering algorithm, wherein the number of clusters is determined by using the elbow method.

10. The method for intelligent detection of surface cracks of a laminated plate according to claim 1, characterized in that: Also includes: The step of performing Gaussian filtering and grayscale processing on the surface image.

Images