Intelligent perception and automatic detection method and system

Abstract

The invention relates to the technical field of flaw detection, in particular to an intelligent perception and automatic detection method and system. Comprising the following steps: S1, obtaining a detection image and carrying out preprocessing, including image denoising and contrast enhancement; performing gridding segmentation on the image; clustering defect areas in the image by using a clustering algorithm; determining the distance between core clustering points by using an Euclidean distance formula, and optimizing the number of points in each cluster by adjusting the radius of the cluster; s2, selecting a clustering point with the lowest similarity by using a similarity measurement method according to a clustering result; and selecting three points with the lowest similarity from the clustering points with the lowest similarity. According to the method, the image quality is improved through preprocessing, defects are extracted and predicted by using CNN and LSTM models, the noise influence is reduced, and the detection stability is improved through clustering and triangle construction; priority classification ensures efficient utilization of resources and rapid identification of defect areas.

Description

Technical Field

[0001] The present invention relates to the field of flaw detection technology, and in particular to an intelligent sensing and automated detection method and system. Background Art

[0002] In modern industrial production and maintenance, flaw detection technology is crucial to discover defects inside and on the surface of materials or structures to ensure product quality and safety. Although traditional flaw detection methods, such as ultrasonic testing, X-ray testing, magnetic particle testing, and penetration testing, meet the detection needs, they generally have a core problem: they rely on manual experience and are difficult to automate. This leads to low detection efficiency, high human errors, and difficulty in widespread application in large-scale production.

[0003] With the rapid development of information technology and intelligent technology, intelligent perception and automated detection methods have gradually become a new trend in the field of flaw detection. Intelligent perception technology uses advanced sensors and image processing technology to obtain multi-dimensional information of the detection object in real time and accurately. The automated detection system uses intelligent algorithms and machine learning technology to achieve efficient processing and analysis of detection data, significantly improving the accuracy and efficiency of detection.

[0004] In the fields of aerospace, automobile manufacturing, energy and electricity, the demand for high-precision, real-time, and automated flaw detection is becoming increasingly urgent. Intelligent perception and automated detection methods can not only improve the accuracy and reliability of detection, but also reduce production costs and ensure product quality and safety. Therefore, researching and developing an intelligent perception and automated detection method and system has important theoretical and practical significance. Summary of the invention

[0005] In order to overcome the shortcomings of relying on manual experience and difficulty in achieving automation, the present invention provides an intelligent perception and automated detection method and system.

[0006] The technical solution of the present invention is: a smart perception and automatic detection method, comprising the following steps: S1: Obtain the inspection image and perform preprocessing, including image denoising and contrast enhancement; gridding and segmenting the image; clustering the defect areas in the image using a clustering algorithm; using the Euclidean distance formula to determine the distance between core cluster points, and optimizing the number of points in each cluster by adjusting the cluster radius; S2: Based on the clustering results, the cluster point with the lowest similarity is selected using the similarity measurement method; three points with the lowest similarity are selected from the cluster points with the lowest similarity to form a triangle, a total of N triangles are generated, and the area with the most intersections is determined; these areas are prioritized according to the number of intersections; S3: Determine the priority analysis area of ​​the image based on the priority classification results; conduct in-depth analysis on the priority analysis area to determine the final defect area; the system automatically adjusts and optimizes the parameters of the image analysis and defect prediction model by continuously collecting data during the inspection process; each inspection result is fed back to the system, forming a continuous self-optimization cycle.

[0007] Preferably, the application of a clustering algorithm to cluster defect areas in an image comprises: first acquiring a preprocessed image; secondly using a CNN model to analyze the preprocessed image and extract defect areas in the image; finally using an LSTM model to predict the probability of defect areas appearing in the image in the future based on time series data; extracting key feature points from the location and confidence of potential defect areas identified by the CNN model and the future defect probability predicted by the LSTM model; describing the extracted feature points and generating feature vectors; selecting a hierarchical clustering algorithm, inputting the feature points and their feature vectors into the hierarchical clustering algorithm, performing clustering analysis, and grouping similar feature points into the same cluster.

[0008] Preferably, the method of using the Euclidean distance formula to determine the distance between core clustering points and optimizing the number of points in each cluster by adjusting the clustering radius includes: first, selecting a core clustering point A among the K clusters, where the core clustering point A is the point with the highest density in the cluster; secondly, taking the core clustering point A as the center, selecting two other core clustering points B and C; the criterion for selecting B and C is to minimize the sum of the distances between the three core points A, B and C; and using the Euclidean distance formula to determine the distance from the core clustering point A to the core clustering points B and C; the Euclidean distance formula is:

[0009] in, is the distance between two cluster core points, ( , )and( , ) are the positions of the two core clustering points respectively; finally, by adjusting the clustering radius, the number of points in each cluster is optimized and it is ensured that there is no overlapping area among the three clustering areas.

[0010] Preferably, the method of obtaining the cluster point with the lowest similarity according to the clustering result based on the similarity measurement includes: obtaining the final clustering result data and extracting each cluster point in the clustering result; first, using the distance correlation coefficient Calculate the similarity between each pair of cluster points; the distance correlation coefficient formula is as follows:

[0011] in, express and The distance correlation coefficient between express and The distance covariance between and Respectively and The distance variance is ; the cluster point with the lowest similarity is selected from all cluster points, that is, the cluster point with the smallest distance correlation coefficient; when all cluster points are analyzed, the maximum information coefficient is used The remaining cluster points are further analyzed to determine the cluster point with the lowest similarity; the maximum information coefficient formula is as follows,

[0012] in, express and The maximum information coefficient between is the maximum number of mesh divisions, Represents the mutual information under a specific grid partitioning.

[0013] Preferably, the method comprises: extracting all points in each cluster after obtaining the final clustering result data; calculating the similarity between each pair of cluster points using a similarity measurement formula; sorting all point pairs according to the calculated similarity values, from the least similar to the most similar; and first selecting a pair of points with the lowest similarity based on the sorting result, and then selecting a point with the lowest similarity to the first two points from the unselected points, so that these three points form a triangle. Preferably, N triangles are generated according to the aforementioned method, and the area with the most intersections is determined; these areas are prioritized according to the number of intersections, including: from the N generated triangles, in the order of triangle generation, from the first triangle to the last triangle; marking the intersection area of ​​each triangle with other triangles in turn, and recording the number of intersections of each intersection area, the number of intersections refers to the number of triangles covering each area; counting the number of intersections of each intersection area, and recording the number of triangles covering each area; classifying the intersection areas according to the number of intersections, the areas with a small number of intersections are low priority; the areas with a medium number of intersections are medium priority; the areas with a large number of intersections are high priority.

[0014] Preferably, determining the priority analysis area of ​​the image according to the priority classification result includes: according to the above classification results, the priority analysis areas of the image are in the order of high priority analysis area, medium priority analysis area, and low priority analysis area; analyzing the high priority analysis area, if obvious defect features are detected in the high priority analysis area, determining the area to be a defect area; if no obvious defect features are detected in the high priority analysis area, turning to the medium priority analysis area for analysis; analyzing the medium priority analysis area, if defect features are detected in the medium priority analysis area, determining whether it is a defect based on the change in the area of ​​the area; if the area of ​​the medium priority analysis area changes in continuous detection, determining the area to be a defect area; if the area of ​​the medium priority analysis area remains stable, further analyzing the contrast features of adjacent areas and the contrast features of historical data to determine whether it is a defect; analyzing the low priority analysis area, if defect features are detected in the low priority analysis area, determining the area to be a suspected defect area.

[0015] Preferably, the in-depth analysis of the priority analysis area to determine the final defect area includes: obtaining the image area of ​​the area to be detected and defining it as a first detection area, then obtaining the defect area around the area to be detected, the area around the area to be detected refers to an area adjacent to the area to be detected and defined as a second detection area, and using the following formula to calculate the defect probability of the suspected defect area, the formula is as follows,

[0016] in, is the first detection area, is the second detection area, is the minimum value; is the defect probability.

[0017] Preferably, the system automatically adjusts and optimizes the parameters of the image analysis and defect prediction model by continuously collecting data during the detection process; each detection result is fed back to the system to form a continuous self-optimization cycle, including: based on the results of clustering and triangle generation, selecting representative triangles, randomly sampling to generate new triangles, calculating the number of intersections and areas, and continuously iterating and optimizing.

[0018] An intelligent perception and automatic detection system, comprising: Image acquisition and preprocessing module: responsible for acquiring the detection image and performing preprocessing, including denoising, contrast enhancement and grid segmentation, to improve image quality; Feature extraction and clustering module: Use the CNN model to extract image features and use LSTM to predict the probability of future defect areas. The generated feature vectors are clustered using a hierarchical clustering algorithm to group similar feature points into the same cluster and determine core points and non-core points. Similarity measurement and triangle construction module: calculate the similarity between cluster points, select the three most dissimilar points to form a triangle, generate multiple triangles and mark the intersection area, and record the number of intersections in each area; Priority analysis area determination module: Classify areas according to the number of intersections, use the reverse strategy to determine the priority, further analyze the defect characteristics in each priority area, and evaluate the defects by combining the comparison of adjacent areas and historical data comparison characteristics; Deep learning model module: Uses CNN and LSTM models to identify and predict defect areas in images, extract key feature points and generate feature vectors, providing accurate defect detection and prediction functions; System optimization and control module: The system continuously collects test data, automatically adjusts and optimizes the parameters of the image analysis and defect prediction models, forms a closed-loop optimization cycle, and controls the operating status and resource utilization of the system.

[0019] Beneficial effects: First, by preprocessing the detected image, including image denoising and contrast enhancement, the image quality is significantly improved, laying a solid foundation for subsequent analysis. Then, the CNN model is used to analyze the preprocessed image and extract the defect area in the image, ensuring the comprehensiveness and accuracy of feature extraction; at the same time, the LSTM model is used to predict the defect area that will appear in the future, enhancing the foresight and reliability of the detection. Subsequently, by calculating the similarity of the clustering points, the three points with the lowest similarity are selected to form a triangle using the distance correlation coefficient and the maximum information coefficient, which effectively reduces the influence of noise and abnormal points and improves the stability and representativeness of the detection. On this basis, the regions are prioritized according to the number of intersections to ensure efficient use of resources and rapid identification of obvious defect areas; high-priority analysis areas are directly determined as defect areas, medium-priority analysis areas are further analyzed according to area changes, and low-priority analysis areas are marked as suspected defect areas for re-inspection or manual review to ensure the comprehensiveness and accuracy of the detection. By continuously optimizing the detection process, adjusting the clustering radius and iterative calculation, the detection accuracy and efficiency are continuously improved, and detailed detection reports are generated, providing reliable guarantees for practical applications. Ultimately, the present invention not only improves the accuracy and reliability of detection, but also significantly improves detection efficiency, reduces human errors, and realizes the automation of the detection process. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flow chart of an intelligent perception and automated detection method and system of the present invention; Figure 2 This is a structural schematic diagram of an intelligent perception and automatic detection system of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Embodiment 1: A smart sensing and automated detection method, such as Figure 1 As shown, the following steps are included: S1: Obtain the inspection image and perform preprocessing, including image denoising and contrast enhancement; gridding and segmenting the image; clustering the defect areas in the image using a clustering algorithm; using the Euclidean distance formula to determine the distance between core cluster points, and optimizing the number of points in each cluster by adjusting the cluster radius; S2: Based on the clustering results, the cluster point with the lowest similarity is selected using the similarity measurement method; three points with the lowest similarity are selected from the cluster points with the lowest similarity to form a triangle, a total of N triangles are generated, and the area with the most intersections is determined; these areas are prioritized according to the number of intersections; S3: Determine the priority analysis area of ​​the image based on the priority classification results; conduct in-depth analysis on the priority analysis area to determine the final defect area; the system automatically adjusts and optimizes the parameters of the image analysis and defect prediction model by continuously collecting data during the inspection process; each inspection result is fed back to the system, forming a continuous self-optimization cycle.

[0023] A clustering algorithm is applied to cluster defective areas in an image, including: first, obtaining a preprocessed image; second, using a CNN model to analyze the preprocessed image and extract defective areas in the image; finally, using an LSTM model to predict the probability of defective areas appearing in the image in the future based on time series data; extracting key feature points from the location and confidence of potential defective areas identified by the CNN model and the future defect probability predicted by the LSTM model; describing the extracted feature points and generating feature vectors; selecting a hierarchical clustering algorithm, inputting the feature points and their feature vectors into the hierarchical clustering algorithm, performing clustering analysis, and classifying similar feature points into the same cluster.

[0024] Further explanation is that the preprocessing includes image denoising and contrast enhancement operations; the feature points are extracted from key positions of the center point and edge point of the defect area; the feature vector includes position information and shape information; by combining CNN and LSTM models, the defect information in the image is captured more comprehensively and the accuracy of detection is improved; the CNN model is good at processing features in static images, while the LSTM model can process time series data and predict defect areas that will appear in the future. The combination of the two provides more comprehensive detection results. Through multi-level feature extraction and clustering, defect areas can be more accurately identified and classified, reducing false alarms and missed alarms; the hierarchical clustering algorithm can handle clusters of different sizes and shapes to enhance the robustness of the system.

[0025] The Euclidean distance formula is used to determine the distance between core clustering points, and the number of points in each cluster is optimized by adjusting the clustering radius, including: first, among the K clusters, a core clustering point A is selected, and the core clustering point A is the point with the highest density in the cluster; secondly, with the core clustering point A as the center, two other core clustering points B and C are selected; the standard for selecting B and C is to minimize the sum of the distances between the three core points A, B and C; the Euclidean distance formula is used to determine the distance from the core clustering point A to the core clustering points B and C; the Euclidean distance formula is:

[0026] in, is the distance between two cluster core points, ( , )and( , ) are the positions of the two core clustering points respectively; finally, by adjusting the clustering radius, the number of points in each cluster is optimized and it is ensured that there is no overlapping area among the three clustering areas.

[0027] A further explanation is that, assuming there are the following point sets, Cluster 1: (1, 1), (1, 2), (2, 1), (2, 2), Cluster 2: (4, 4), (4, 5), (5, 4), (5, 5), Cluster 3: (7, 7), (7, 8), (8, 7), (8, 8) Select the core cluster point A, select the core cluster point A from cluster 1, and assume that point A is (1, 1) because it is the point with the highest density. Select core cluster points B and C, select core cluster points B and C from cluster 2 and cluster 3 so that (d(A, B)+d(B, C)+d(A, C)) is minimized. Assume that B is selected as (4, 4) and C is selected as (7, 7). Calculate the distance between the core cluster points and calculate (d(A, B)):

[0028] ; Calculate (d(B, C)):

[0029] ; Calculate (d(A, C)):

[0030] ; Ensure that the sum of distances between the three core points is minimal, total distance:

[0031] ; Adjust the cluster radius to ensure that the three cluster areas have no intersection, and gradually reduce the cluster radius to ensure that each cluster area no longer overlaps. For example, set the radius of cluster 1 to 1.5, the radius of cluster 2 to 1.5, and the radius of cluster 3 to 1.5, so that the three cluster areas will not have any intersection.

[0032] According to the clustering results, the clustering points with the lowest similarity are obtained according to the similarity measurement, including: obtaining the final clustering result data and extracting each clustering point in the clustering result; first, using the distance correlation coefficient (dCor(X, Y)) to calculate the similarity between each pair of clustering points; wherein the distance correlation coefficient formula is as follows, in, express and The distance correlation coefficient between express and The distance covariance between and Respectively and The distance variance is ; the cluster point with the lowest similarity is selected from all cluster points, that is, the cluster point with the smallest distance correlation coefficient; when all cluster points are analyzed, the maximum information coefficient is used The remaining cluster points are further analyzed to determine the cluster point with the lowest similarity; the maximum information coefficient formula is as follows,

[0033] in, express and The maximum information coefficient between is the maximum number of mesh divisions, Represents the mutual information under a specific grid partitioning.

[0034] A further explanation is that the distance correlation coefficient : Robustness: The distance correlation coefficient can detect both linear and nonlinear relationships and is applicable to all types of data; Standardization: The value range of the distance correlation coefficient is between 0 and 1, which is convenient for comparing the similarities between different variables; Zero mean: The distance correlation coefficient can work effectively even if the data does not have a zero mean; Maximum information coefficient : Detecting complex relationships: The maximum information coefficient can detect various complex nonlinear relationships, including periodicity, monotonicity, and other complex patterns; Interpretability: The value range of the maximum information coefficient is between 0 and 1. The closer the value is to 1, the stronger the relationship between the two variables; Flexibility: By adjusting the maximum number of grid divisions, it can adapt to data sets of different sizes; Suppose there are the following cluster points: Cluster 1: (1, 1), (1, 2), (2, 1), (2, 2), Cluster 2: (4, 4), (4, 5), (5, 4), (5, 5), Cluster 3: (7, 7), (7, 8), (8, 7), (8, 8) Calculate the distance correlation coefficient: 1. Select a pair of cluster points (1, 1) and (4, 4): Calculate :

[0035] ;calculate and :

[0036]

[0037] ;calculate :

[0038] ; 2. Select the cluster point with the lowest similarity: Calculate the distance correlation coefficient between all cluster points. Select a pair of cluster points with the smallest distance correlation coefficient, such as (1, 1) and (4, 4); Calculate the maximum information coefficient 1. Select the remaining cluster points (1, 1) and (7, 7): Calculate the mutual information under different grid divisions ; Select the maximum number of mesh divisions ; Calculate using the formula :

[0039] ; 2. Select the cluster point with the lowest similarity: calculate the maximum information coefficient between all remaining cluster points; select a pair of cluster points with the smallest maximum information coefficient, such as (1, 1) and (7, 7).

[0040] According to the clustering results, the cluster point with the lowest similarity is selected by using the similarity measurement method; the three points with the lowest similarity are selected from the cluster points with the lowest similarity to form a triangle, including: after obtaining the final clustering result data, all points in each cluster are extracted; the similarity between each pair of cluster points is calculated using the similarity measurement formula; all point pairs are sorted from the least similar to the most similar according to the calculated similarity values; based on the sorting results, a pair of points with the lowest similarity is first selected, and then a point with the lowest similarity to the first two points is selected from the unselected points, and these three points form a triangle. Further explanation is that by using the distance correlation coefficient and the maximum information coefficient, various types of similarity relationships can be detected to improve the robustness of the method; the three points with the lowest similarity are selected to ensure their independence, thereby improving the stability and representativeness of the triangle; by selecting the point pair with the lowest similarity, the influence of noise and abnormal points is reduced, and the accuracy of detection is improved; the value range of the similarity measurement formula is between 0 and 1, which is easy to understand and compare.

[0041] Generate N triangles according to the above method, and determine the area with the most intersections; prioritize these areas according to the number of intersections, including: from the N generated triangles, in the order of triangle generation, from the first triangle to the last triangle; mark the intersection area of ​​each triangle with other triangles in turn, and record the number of intersections of each intersection area, the number of intersections refers to how many triangles cover each area; count the number of intersections of each intersection area, and record how many triangles cover each area; classify the intersection areas according to the number of intersections, the areas with a small number of intersections are low priority; the areas with a medium number of intersections are medium priority; the areas with a large number of intersections are high priority.

[0042] For further explanation, assume that there are the following 5 triangles: T1: vertices (1, 1), (2, 2), (3, 1), T2: vertices (2, 2), (3, 3), (4, 2), T3: vertices (3, 1), (4, 2), (5, 1), T4: vertices (4, 2), (5, 3), (6, 2), T5: vertices (5, 1), (6, 2), (7, 1), detailed steps 1. Generate N triangles: 5 triangles have been generated: T1, T2, T3, T4, T5; 2. Mark the intersection area of ​​each triangle with other triangles: Mark the intersection area of ​​each triangle with other triangles in turn, and record the number of intersections of each intersection area; T1 and T2: intersection Set area R1, number of intersections + 1; T1 and T3: intersection area R2, number of intersections + 1; T2 and T3: intersection area R3, number of intersections + 1; T2 and T4: intersection area R4, number of intersections + 1; T3 and T4: intersection area R5, number of intersections + 1; T3 and T5: intersection area R6, number of intersections + 1; T4 and T5: intersection area R7, number of intersections + 1; 3. Count the number of intersections of each intersection area: R1: number of intersections 1, R2: number of intersections 1, R3: number of intersections 1, R4: number of intersections 1, R5: number of intersections 1, R6: number of intersections 1, R7: number of intersections 1; 4. Classify according to the number of intersections: Low priority: areas with fewer intersections. In the current example, all intersection regions (R1, R2, R3, R4, R5, R6, and R7) have 1 intersection, so they are all classified as low priority. Medium and high priority: In the current example, there are no regions with more than 1 intersection, so these categories are not applicable yet. If there are more triangles or more complex intersections in the future, and more intersections occur, they will be further subdivided into medium priority (regions with a medium number of intersections) and high priority (regions with a large number of intersections).

[0043] According to the priority classification result, the priority analysis area of ​​the image is determined, including: according to the above classification results, the priority analysis areas of the image are in the order of high priority analysis area, medium priority analysis area, and low priority analysis area; analyzing the high priority analysis area, if obvious defect features are detected in the high priority analysis area, the area is determined to be a defect area; if no obvious defect features are detected in the high priority analysis area, turn to the medium priority analysis area for analysis; analyzing the medium priority analysis area, if defect features are detected in the medium priority analysis area, determine whether it is a defect based on the change in the area of ​​the area; if the area of ​​the medium priority analysis area changes in continuous detection, determine the area to be a defect area; if the area of ​​the medium priority analysis area remains stable, further analyze the contrast features of adjacent areas and the contrast features of historical data to determine whether it is a defect; analyzing the low priority analysis area, if defect features are detected in the low priority analysis area, determine the area to be a suspected defect area.

[0044] Further explanation is that the obvious defect features include but are not limited to color changes and texture anomalies; low priority analysis areas are areas with the least number of intersections; although these areas have the least number of intersections, they are the most stable and also hide defects that are not easy to find; if defect features are detected, they are determined to be suspected defect areas for re-inspection or manual review; medium priority analysis areas are areas with a medium number of intersections; these areas are stable but still need attention. If defect features are detected, it is necessary to further analyze the changes in the area of ​​the area; if the area of ​​the area changes during continuous detection, it is considered that there is a defect in the area; if the area of ​​the area remains stable, other features are further analyzed to determine whether it is a defect; high priority analysis areas are areas with the most intersections; these areas are the most unstable, but also the easiest places to find problems; if obvious defect features are detected, they are directly determined as defect areas; quickly identify obvious defect areas to reduce the workload of subsequent analysis.

[0045] Conduct an in-depth analysis of the priority analysis area to determine the final defect area, including: obtaining the image area of ​​the area to be detected and defining it as the first detection area, then obtaining the defect area around the area to be detected, the area around the area to be detected refers to the area adjacent to the area to be detected and defined as the second detection area, and using the following formula to calculate the defect probability of the suspected defect area, the formula is as follows,

[0046] in, is the first detection area, is the second detection area, is the minimum value; is the defect probability.

[0047] Further explanation is that the parameters in the formula and The total area of ​​the region is calculated, that is, not just the defect area in the region; assuming The area of ​​the region is 10, and the suspected defect area is 2. The area of ​​the region is 20 and the defect area is 6, so the formula calculates is 10, is 20.

[0048] The system automatically adjusts and optimizes the parameters of the image analysis and defect prediction models by continuously collecting data during the inspection process. Each inspection result is fed back to the system, forming a continuous self-optimization cycle, including: selecting representative triangles based on the results of clustering and triangle generation, randomly sampling to generate new triangles, calculating the number of intersections and areas, and continuously iterating and optimizing.

[0049] Example 2: Based on Example 1, an intelligent perception and automatic detection system includes.

[0050] Image acquisition and preprocessing module: responsible for acquiring the detection image and performing preprocessing, including denoising, contrast enhancement and grid segmentation, to improve image quality; Feature extraction and clustering module: Use the CNN model to extract image features and use LSTM to predict the probability of future defect areas. The generated feature vectors are clustered using a hierarchical clustering algorithm to group similar feature points into the same cluster and determine core points and non-core points. Similarity measurement and triangle construction module: calculate the similarity between cluster points, select the three most dissimilar points to form a triangle, generate multiple triangles and mark the intersection area, and record the number of intersections in each area; Priority analysis area determination module: Classify areas according to the number of intersections, use the reverse strategy to determine the priority, further analyze the defect characteristics in each priority area, and evaluate the defects by combining the comparison of adjacent areas and historical data comparison characteristics; Deep learning model module: Uses CNN and LSTM models to identify and predict defect areas in images, extract key feature points and generate feature vectors, providing accurate defect detection and prediction functions; System optimization and control module: The system continuously collects test data, automatically adjusts and optimizes the parameters of the image analysis and defect prediction models, forms a closed-loop optimization cycle, and controls the operating status and resource utilization of the system.

[0051] The above is a detailed introduction to the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A smart perception and automated detection method, characterized in that: The following steps are involved: S1: Acquire the detection image and perform preprocessing, including image denoising and contrast enhancement; Image grid segmentation; clustering algorithm is applied to cluster defect areas in the image; the distance between core cluster points is determined by using Euclidean distance formula, and the number of points in each cluster is optimized by adjusting the cluster radius; S2: Based on the clustering results, the cluster point with the lowest similarity is selected using the similarity measurement method; three points with the lowest similarity are selected from the cluster points with the lowest similarity to form a triangle, a total of N triangles are generated, and the area with the most intersections is determined; these areas are prioritized according to the number of intersections; S3: Determine the priority analysis area of ​​the image based on the priority classification results; conduct in-depth analysis on the priority analysis area to determine the final defect area; the system automatically adjusts and optimizes the parameters of the image analysis and defect prediction model by continuously collecting data during the inspection process; each inspection result is fed back to the system, forming a continuous self-optimization cycle.

2. The intelligent perception and automatic detection method according to claim 1, characterized in that: The method of applying a clustering algorithm to cluster defect areas in an image includes: firstly obtaining a preprocessed image; secondly using a CNN model to analyze the preprocessed image and extracting defect areas in the image; finally using an LSTM model to predict the probability of defect areas appearing in the image in the future based on time series data; extracting key feature points from the position and confidence of the potential defect area identified by the CNN model and the future defect probability predicted by the LSTM model; describing the extracted feature points and generating feature vectors; selecting a hierarchical clustering algorithm, inputting the feature points and their feature vectors into the hierarchical clustering algorithm, performing clustering analysis, and classifying similar feature points into the same cluster.

3. According to the intelligent perception and automatic detection method of claim 1, it is characterized in that: The method uses the Euclidean distance formula to determine the distance between core clustering points, and optimizes the number of points in each cluster by adjusting the clustering radius, including: first, selecting a core clustering point A among the K clusters, where the core clustering point A is the point with the highest density in the cluster; second, taking the core clustering point A as the center, selecting another two core clustering points B and C; the criterion for selecting B and C is to minimize the sum of the distances between the three core points A, B and C; and using the Euclidean distance formula to determine the distance from the core clustering point A to the core clustering points B and C; the Euclidean distance formula is: in, is the distance between two cluster core points, ( , )and( , ) are the positions of the two core clustering points respectively; finally, by adjusting the clustering radius, the number of points in each cluster is optimized and it is ensured that there is no overlapping area among the three clustering areas.

4. According to the intelligent perception and automatic detection method of claim 1, it is characterized in that: According to the clustering results, the clustering point with the lowest similarity is obtained according to the similarity measurement, including: obtaining the final clustering result data and extracting each clustering point in the clustering result; first, using the distance correlation coefficient Calculate the similarity between each pair of cluster points; the distance correlation coefficient formula is as follows: in, express and The distance correlation coefficient between express and The distance covariance between and Respectively and The distance variance is ; the cluster point with the lowest similarity is selected from all cluster points, that is, the cluster point with the smallest distance correlation coefficient; when all cluster points are analyzed, the maximum information coefficient is used The remaining cluster points are further analyzed to determine the cluster point with the lowest similarity; the maximum information coefficient formula is as follows, in, express and The maximum information coefficient between is the maximum number of mesh divisions, Represents the mutual information under a specific grid partitioning.

5. According to the intelligent perception and automatic detection method of claim 1, it is characterized in that: The method comprises: extracting all points in each cluster after obtaining the final clustering result data; calculating the similarity between each pair of cluster points using a similarity measurement formula; sorting all point pairs according to the calculated similarity values, from the least similar to the most similar; and first selecting a pair of points with the lowest similarity based on the sorting result, and then selecting a point with the lowest similarity to the first two points from the unselected points, so that the three points form a triangle.

6. The intelligent perception and automatic detection method according to claim 1, characterized in that: The method of generating N triangles according to the above method and determining the area with the largest number of intersections; Prioritize these areas according to the number of intersections, including: from the N generated triangles, in the order in which the triangles are generated, from the first triangle to the last triangle; Mark the intersection areas of each triangle with other triangles in turn, and record the number of intersections of each intersection area. The number of intersections refers to how many triangles cover each area; count the number of intersections of each intersection area, and record how many triangles cover each area; classify the intersection areas according to the number of intersections, and the areas with a small number of intersections have a low priority; the areas with a medium number of intersections have a medium priority; and the areas with a large number of intersections have a high priority.

7. A smart perception and automatic detection method according to claim 6, characterized in that: The method of determining the priority analysis area of ​​the image according to the priority classification result includes: according to the above classification result, the priority analysis areas of the image are in the order of high priority analysis area, medium priority analysis area, and low priority analysis area; analyzing the high priority analysis area, if obvious defect features are detected in the high priority analysis area, determining the area as a defect area; if no obvious defect features are detected in the high priority analysis area, turning to the medium priority analysis area for analysis; analyzing the medium priority analysis area, if defect features are detected in the medium priority area, determining whether it is a defect based on the change in the area of ​​the area; if the area of ​​the medium priority analysis area changes in continuous detection, determining the area as a defect area; if the area of ​​the medium priority analysis area remains stable, further analyzing the contrast features of adjacent areas and the contrast features of historical data to determine whether it is a defect; analyzing the low priority analysis area, if defect features are detected in the high priority analysis area, determining the area as a suspected defect area.

8. The intelligent perception and automatic detection method according to claim 1, characterized in that: The in-depth analysis of the priority analysis area to determine the final defect area includes: obtaining the image area of ​​the area to be detected and defining it as a first detection area, then obtaining the defect area of ​​the area around the area to be detected and adjacent to the area to be detected, the area around the area to be detected refers to the area adjacent to the area to be detected and is defined as a second detection area, and the defect probability of the suspected defect area is calculated using the following formula, which is as follows: in, is the first detection area, is the second detection area, is the minimum value; is the defect probability.

9. The intelligent perception and automatic detection method according to claim 1, characterized in that: The system automatically adjusts and optimizes the parameters of the image analysis and defect prediction models by continuously collecting data during the inspection process; Each detection result is fed back to the system, forming a continuous self-optimization cycle, including: selecting representative triangles based on the results of clustering and triangle generation, randomly sampling to generate new triangles, calculating the number of intersections and areas, and continuously iterating and optimizing.

10. An intelligent perception and automatic detection system, used in the intelligent perception and automatic detection method according to any one of claims 1 to 9, characterized in that: Included are: Image acquisition and preprocessing module: responsible for acquiring the detection image and performing preprocessing, including denoising, contrast enhancement and grid segmentation, to improve image quality; Feature extraction and clustering module: Use the CNN model to extract image features and use LSTM to predict the probability of future defect areas. The generated feature vectors are clustered using a hierarchical clustering algorithm to group similar feature points into the same cluster and determine core points and non-core points. Similarity measurement and triangle construction module: calculate the similarity between cluster points, select the three most dissimilar points to form a triangle, generate multiple triangles and mark the intersection area, and record the number of intersections in each area; Priority analysis area determination module: Classify areas according to the number of intersections, use the reverse strategy to determine the priority, further analyze the defect characteristics in each priority area, and evaluate the defects by combining the comparison of adjacent areas and historical data comparison characteristics; Deep learning model module: Uses CNN and LSTM models to identify and predict defect areas in images, extract key feature points and generate feature vectors, providing accurate defect detection and prediction functions; System optimization and control module: The system continuously collects test data, automatically adjusts and optimizes the parameters of the image analysis and defect prediction models, forms a closed-loop optimization cycle, and controls the operating status and resource utilization of the system.

Images