An industrial component defect detection method and system based on X-ray

Abstract

The present invention provides an industrial component defect detection method and system based on X-rays, which relates to the technical field of material analysis. The method comprises: obtaining an X-ray image of an industrial component to be detected; extracting color features and texture features of the X-ray image; constructing a feature vector according to the extracted color features and texture features; determining the abnormal value of each pixel point according to the feature vector of each pixel point; and performing defect detection on the industrial component to be detected based on threshold segmentation according to the abnormal value of each pixel point, and determining the defect position of the industrial component to be detected. The present invention does not require load training, has a simple calculation process, saves time and has low calculation cost, is conducive to batch detection of industrial components, and has high defect detection efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of material analysis, and in particular to an industrial component defect detection method and system based on X-rays. Background Art

[0002] In industrial production processes, automated detection of component defects has always been an important and challenging problem.

[0003] X-ray inspection is a non-destructive inspection method that can detect internal defects without damaging the component. It can reveal internal defects such as cracks, holes, bubbles and foreign matter that may not be detected during the production process. By promptly detecting and repairing these defects, product quality can be significantly improved and product returns and recalls caused by defects can be reduced.

[0004] With the rapid development of technologies such as pattern recognition, machine vision, and deep learning, as well as the urgent need for automated detection of component defects, more and more modern technologies are being applied to automated detection of component defects, such as the use of YOLO and SSD neural network target detection algorithms for automated detection of component defects.

[0005] However, current automated inspection technologies based on neural networks require a large amount of well-labeled data for training. Acquiring and labeling this data is usually expensive and time-consuming. At the same time, a large amount of computing resources is required during the inspection process, which is time-consuming and computationally expensive, which is not conducive to batch inspection of industrial parts and has low defect detection efficiency. Summary of the invention

[0006] In order to solve the existing problems of the current automatic detection technology based on neural network in the prior art, a large amount of well-labeled data is required for training. It is usually expensive and time-consuming to obtain and label this data. At the same time, a large amount of computing resources is required in the detection process, which is time-consuming and has high computing costs, is not conducive to batch detection of industrial parts, and has low defect detection efficiency. The present invention provides a method and system for industrial parts defect detection based on X-rays.

[0007] The technical solution provided by the embodiment of the present invention is as follows:

[0008] First aspect:

[0009] An embodiment of the present invention provides an X-ray-based industrial component defect detection method, comprising:

[0010] S1: Acquire an X-ray image of the industrial component to be inspected;

[0011] S2: extracting color features and texture features of the X-ray image;

[0012] S3: construct a feature vector based on the extracted color features and texture features;

[0013] S4: Determine the outlier value of each pixel point according to the feature vector of each pixel point;

[0014] S5: According to the abnormal value of each pixel point, based on the threshold segmentation, defect detection is performed on the industrial component to be detected, and the defect position of the industrial component to be detected is determined.

[0015] Second aspect:

[0016] An embodiment of the present invention provides an X-ray-based industrial component defect detection system, comprising:

[0017] processor;

[0018] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the X-ray-based industrial component defect detection method as described in the first aspect is implemented.

[0019] The third aspect:

[0020] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method for industrial component defect detection based on X-rays as described in the first aspect is implemented.

[0021] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0022] In the present invention, non-destructive testing based on X-rays is adopted to extract color features and texture features of X-ray images, and the abnormal values ​​of each pixel in the image are calculated based on the color features and texture features, and then based on threshold segmentation, the defect position of the industrial component to be inspected is determined. There is no need for load training, the calculation process is simple, time is saved, and the calculation cost is low, which is conducive to batch detection of industrial components and has high defect detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic diagram of a process flow of an industrial component defect detection method based on X-rays provided in an embodiment of the present invention;

[0025] Figure 2A schematic structural diagram of an industrial component defect detection system based on X-rays provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0027] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0028] Reference Manual Attached Figure 1 , showing a flow chart of an X-ray-based industrial component defect detection method provided in an embodiment of the present invention.

[0029] The embodiment of the present invention provides an X-ray-based industrial component defect detection method, which can be implemented by an X-ray-based industrial component defect detection device, which can be a terminal or a server. The processing flow of the X-ray-based industrial component defect detection method can include the following steps:

[0030] S1: Acquire an X-ray image of the industrial component to be inspected.

[0031] S2: Extract color features and texture features of X-ray images.

[0032] Among them, the color features include: contrast, entropy value and grayscale average.

[0033] It should be noted that contrast is a measure that describes the degree of difference between areas of different brightness in an image. In a high-contrast image, the difference between the bright and dark areas is large, and the edges are clear; low-contrast images appear gray and lack a sense of hierarchy. Entropy is a statistic that measures the amount of information and complexity of an image, reflecting the randomness and uncertainty of the grayscale distribution of the image. The higher the entropy value, the more complex and detailed the image is; the lower the entropy value, the simpler and more uniform the image is. Grayscale mean is a basic statistic that describes image brightness, indicating the average level of pixel grayscale values ​​in an image or local area.

[0034] Optionally, the color feature extraction method specifically includes:

[0035] Get the grayscale histogram of each pixel in the X-ray image under the local window.

[0036] Determine the occurrence probability of each gray level in the local window.

[0037] According to the gray level of each pixel in the local window and the probability of occurrence of the gray level, the contrast, entropy value and gray average value of each pixel are calculated:

[0038]

[0039]

[0040]

[0041] in, con i Indicates i The contrast ratio at each pixel, E i Indicates i The entropy value at the pixel point is mean i Indicates i The average grayscale value at each pixel is g represents the gray level, P i ( g ) indicates that each gray level is in the i The probability of occurrence in a local window centered on a pixel point, G represents the grayscale value set, N g Indicates the maximum gray level.

[0042] In the present invention, in the X-ray image, by calculating the contrast, entropy value and grayscale average value in the local window, the local features of the image can be effectively captured and described. This method has significant advantages in defect detection, image segmentation and classification tasks, and can improve the accuracy and robustness of detection.

[0043] Among them, the texture features include: local kernel regression texture features.

[0044] Optionally, the texture feature extraction method specifically includes:

[0045] The texture features of X-ray images are extracted by local kernel regression:

[0046]

[0047] in, F i Indicates i The local kernel regression texture features at the pixel point, exp represents the exponential function with the natural constant as the base, X j Indicates the first j The position coordinates of the pixel points, X i Indicates the first i The position coordinates of the pixel points, ( ) T represents the matrix transpose operation, Cj Indicates the first j The gradient covariance matrix at the pixel point is , n Represents the total number of pixels within the local window.

[0048] The gradient covariance matrix is ​​specifically:

[0049]

[0050] in, represents the first-order grayscale derivative along the horizontal direction, represents the first-order grayscale derivative along the vertical direction, Ω i Indicates the i The local window is centered at a pixel point.

[0051] In the present invention, the local kernel regression texture feature extraction method can accurately and robustly describe the local texture structure of the image by combining local gradient information and covariance matrix. This method has significant advantages in noise resistance, stability, adaptability and detection accuracy, and is particularly suitable for industrial component defect detection tasks that require high accuracy and high robustness.

[0052] S3: Construct a feature vector based on the extracted color features and texture features.

[0053] Optionally, the feature vector is specifically:

[0054]

[0055] in, I i Indicates i The feature vector of pixels, con i Indicates i The contrast ratio at each pixel, E i Indicates i The entropy value at the pixel point is mean i Indicates i The average grayscale value at each pixel is F i Indicates i Local kernel regression texture features at pixels.

[0056] S4: Determine the outlier value of each pixel point according to the feature vector of each pixel point.

[0057] In a possible implementation manner, S4 is specifically:

[0058] Determine the outlier value of each pixel according to the following formula:

[0059]

[0060] in, A ( X i ) indicates the i The outlier value at the pixel point, Ω i Indicates the i The local window centered at the pixel point, X i Indicates i The position coordinates of the pixel points, X j Indicates the first j The position coordinates of the pixel points, D Indicates the difference between pixels. Indicates i Pixels and j The difference between pixels, w ij Indicates j Pixels for the i The influence coefficient of each pixel.

[0061] In the present invention, since the information of multiple pixel points in the local window is integrated and the calculation of the outlier value in the local area is smoothed through a dynamic weight adjustment mechanism, the calculation of the outlier value is made more comprehensive and accurate.

[0062] The specific differences are:

[0063]

[0064] in, con i Indicates i The contrast ratio at each pixel, E i Indicates i The entropy value at the pixel point is mean i Indicates i The average grayscale value at each pixel is F i Indicates i The local kernel regression texture feature at each pixel point is con j Indicates j The contrast ratio at each pixel, E j Indicates j The entropy value at the pixel point is mean j Indicates jThe average grayscale value at each pixel is F j Indicates j The local kernel regression texture feature at each pixel point is l 1 represents the weight coefficient of contrast, l 2 The weight coefficient representing the entropy value, l 3 Represents the weight coefficient of the grayscale average value, l 4 Represents the weight coefficient of local kernel regression texture features.

[0065] Among them, those skilled in the art can set the contrast weight coefficient according to actual conditions. l 1 , the weight coefficient of entropy value l 2 , the weight coefficient of the grayscale average l 3 And the weight coefficient of local kernel regression texture feature l 4 The present invention does not limit the size.

[0066] In the present invention, by combining contrast, entropy, grayscale average and local kernel regression texture feature differences, the differences between pixels can be described more comprehensively. This multi-dimensional feature fusion improves the accuracy and reliability of anomaly detection.

[0067] The specific impact coefficient is:

[0068]

[0069] in, w ij Indicates j Pixels for the i The influence coefficient of each pixel is e represents a hyperparameter that prevents the denominator from being zero, D min Indicates the minimum difference within the local window.

[0070] It should be noted that the influence coefficient is dynamically adjusted according to the difference so that pixels with high similarity contribute more to the calculation of outliers. This dynamic weight adjustment mechanism can more accurately capture the structural features and outliers of local areas.

[0071] S5: According to the abnormal value of each pixel point, based on the threshold segmentation, defect detection is performed on the industrial component to be detected, and the defect position of the industrial component to be detected is determined.

[0072] In a possible implementation, S5 specifically includes: when the abnormal value of the pixel point is greater than the preset abnormal value, determining that the position of the current pixel point is a defect position.

[0073] Among them, those skilled in the art can set the size of the preset abnormal value according to actual conditions, and the present invention does not limit it.

[0074] In the present invention, using a preset outlier threshold for judgment is a simple and direct method that can quickly determine whether a pixel is a defective location. No complicated post-processing steps are required, simplifying the entire detection process. Compared with complex classification algorithms or post-processing steps, the use of threshold judgment methods in automated detection systems can achieve high-speed and efficient online detection, meeting the needs of real-time monitoring and detection.

[0075] In a possible implementation, S5 specifically includes: when the abnormal value of the pixel point is greater than the preset abnormal value and the local variance is less than the preset variance value, determining that the position of the current pixel point is a defect position.

[0076] Optionally, the local variance is specified as:

[0077]

[0078] in, Var i Indicates i The local variance at each pixel is X j Indicates the first j The position coordinates of the pixel points, A ( X i ) indicates the j The outlier value at the pixel point, Ω i Indicates the i The local window centered at the pixel point, w ij Indicates j Pixels for the i The influence coefficient of each pixel is Indicates the i The average value of outliers in the local window centered at pixels.

[0079] Among them, those skilled in the art can set the size of the preset abnormal value and the preset variance value according to actual conditions, and the present invention does not limit this.

[0080] In the present invention, not only the outliers of pixels are relied on, but also the local variance is combined, which can reduce the misjudgment caused by a single feature and improve the accuracy of detection. By introducing the judgment of local variance, those pixels with high outliers but large local changes can be filtered out, which are usually not defects, but high outliers caused by normal textures or edges.

[0081] Furthermore, in order to obtain more accurate defect detection parameters, for example, the weight coefficient of contrast l 1 , the weight coefficient of entropy value l 2 , the weight coefficient of the grayscale average l 3 , weight coefficient of local kernel regression texture feature l 4 , preset outliers and preset variance values, etc., and optimization algorithms such as whale optimization algorithm and simulated annealing algorithm can be used to determine the optimal defect detection parameters.

[0082] Specifically, the inverse of the mean square error loss function is used as the fitness function of the optimization algorithm.

[0083] A global search is performed using the whale optimization algorithm to identify the suboptimal solution.

[0084] Among them, the Whale Optimization Algorithm (WOA) is an intelligent optimization algorithm based on the predation behavior of whale groups. The algorithm simulates the predation behavior of humpback whales, including the process of encirclement, spiral bubble net predation and search for prey.

[0085] Initialize the whale population, which contains multiple whale individuals. The position of each whale individual in the search space represents a feasible defect detection parameter.

[0086] Generate a random number between 0 and 1 r 1 , calculate the first coefficient A :

[0087]

[0088] in, a Represents the adaptive convergence factor.

[0089] Optionally, an adaptive convergence factor a Specifically:

[0090]

[0091] in, t Indicates the current iteration number,T Indicates the maximum number of iterations.

[0092] It should be noted that in the initial stage of the algorithm, the adaptive convergence factor is large, which makes the search range wider and allows whales to explore a large range to avoid falling into the local optimum too early. In the later stage of the algorithm, the adaptive convergence factor gradually decreases, narrowing the search range, allowing whales to conduct fine searches in a smaller range and improve the accuracy of the solution.

[0093] when Go to the next step. , enter the random search mechanism.

[0094] Generate random variables p .

[0095] when When , the shrinking and encircling mechanism is entered. When the spiral mechanism is entered.

[0096] Determine the current global optimal position, and other whales move towards the global optimal position, and update their own positions through the shrinking and surrounding mechanism:

[0097]

[0098] in, X ( t +1) indicates the t + Individual whale positions at iteration 1, X best ( t ) indicates the t The global optimal position at the iteration is A represents the first coefficient, D represents the randomized distance between the individual position of the whale and the current global optimal position, C represents the second coefficient, t Indicates the current iteration number, X ( t ) indicates the t The individual whale positions at iteration .

[0099] It should be noted that through the shrinking and encircling mechanism, individual whales can not only explore widely in the search space, but also concentrate on developing the area near the optimal solution. By moving to the current global optimal position, individual whales can approach the global optimal solution faster and accelerate the convergence process of the algorithm. By continuously updating the global optimal position and allowing all individuals to move closer to it, individuals in the population work together, make full use of each other's information, and maintain the diversity of the population.

[0100] Generate a random number between 0 and 1 r2 , the whale updates its position through the spiral mechanism by moving in an upward spiral and continuously shrinking the encirclement:

[0101]

[0102] in, D * represents the distance between the individual position of the whale and the current global optimal position, e represents a natural constant, b represents the logarithmic spiral constant, l Represents a random number between -1 and 1.

[0103] It should be noted that according to the random number r 2 The value of , selects different spiral paths (cosine or sine) to increase path diversity. Updating the individual positions of whales through the spiral mechanism can enhance search capabilities and flexibility, accurately control the convergence process, increase the convergence speed, and enhance the robustness and adaptability of the algorithm. This mechanism enables the whale optimization algorithm to more efficiently find the optimal solution when dealing with complex multi-objective optimization problems, thereby improving production efficiency and product quality.

[0104] Based on the positions of individual whales relative to each other, they update their own positions through a random search mechanism:

[0105]

[0106] in, X rand Represents the position of a randomly selected individual whale.

[0107] It should be noted that by randomly selecting the positions of other whales for updating, we can effectively jump out of the local optimal solution and enhance the global search ability of the algorithm. By randomly selecting the positions of different individuals for updating, each whale individual will move in a different direction, thereby increasing the diversity of the population and helping to explore a larger search space.

[0108] Update the global optimal position.

[0109] Perform a Cauchy mutation perturbation on the global optimal position:

[0110]

[0111] in, represents the global optimal position after the Cauchy mutation perturbation, Cauchy (0,1) represents the Cauchy operator.

[0112] It should be noted that the Cauchy distribution has a long tail, which can produce larger disturbances, thereby expanding the search range and enhancing the global exploration ability of the algorithm. The Cauchy mutation can cause a large random disturbance of the global optimal position, which helps to jump out of the local optimal solution and find a better global solution.

[0113] Determine whether the current number of iterations has reached the maximum number of iterations. If so, output the defect detection parameters of the whale individual with the highest current fitness as the suboptimal solution. Otherwise, return to continue iterating.

[0114] According to the suboptimal solution, the optimal solution is determined as the best defect detection parameter through simulated annealing algorithm.

[0115] Among them, the simulated annealing algorithm (SA) is a probabilistic optimization algorithm based on the simulation of the solid annealing process. In the physical annealing process, the material is heated to a high temperature and then slowly cooled, so that its internal atoms gradually arrange into a minimum energy state, thus forming a crystal structure with minimum energy. The simulated annealing algorithm draws on this process and is used to solve combinatorial optimization problems.

[0116] Specifically, the initial temperature T 0 , maximum number of iterations, termination temperature T m .

[0117] The whale individual with the highest fitness determined by the whale optimization algorithm is used as the current individual, and the neighborhood individuals are generated according to the suboptimal model parameters.

[0118] Compare the fitness values ​​of the current individual with those of the neighboring individuals. When the fitness value of the neighboring individual is greater than the fitness value of the current individual, use the neighboring individual to replace the current individual. When the fitness value of the neighboring individual is less than the fitness value of the current individual, replace the current individual with the replacement probability. P Replace the current individual with a neighboring individual.

[0119] Replacement probability P Specifically:

[0120]

[0121] in, P represents the replacement probability, exp represents the e The exponential function with base , X new represents the neighborhood individuals, F ( X new ) represents the fitness value of the neighborhood individual, X t Indicates tThe current individual at the iteration, F ( X t ) indicates the t The fitness value of the current individual at the iteration, T Indicates the current temperature.

[0122] It should be noted that by setting the replacement probability, the simulated annealing algorithm balances the probability of accepting inferior solutions. This can achieve a balance between global search and local search and prevent the search from being too biased in one direction.

[0123] Determine whether the current number of iterations has reached the maximum number of iterations, or whether the current temperature has reached the termination temperature T m If yes, output the optimal model parameters. Otherwise, update the temperature and continue to generate neighborhood individuals for iteration.

[0124] In a possible implementation, the temperature update method is:

[0125]

[0126] in, α represents the temperature drop coefficient, T t+1 Indicates t The temperature at +1 iteration, T t Indicates t The temperature at the iteration, T 0 Indicates the initial temperature.

[0127] In the present invention, by combining the whale optimization algorithm and the simulated annealing algorithm for optimization, the advantages of the two algorithms can be fully utilized, and both a wide global search can be performed to find a suboptimal solution, and a fine local optimization can be performed on the basis of the suboptimal solution to find the optimal solution. This combined optimization strategy not only improves the global search capability and local optimization accuracy, but also enhances the robustness and stability of the algorithm, improves the optimization efficiency, and is suitable for various complex optimization problems and practical application scenarios, such as the optimization of defect detection parameters.

[0128] Reference Manual Attached Figure 2 , showing a schematic structural diagram of an X-ray-based industrial component defect detection system provided by the present invention.

[0129] The present invention further provides an X-ray based industrial component defect detection system 20, which is applied to the above-mentioned X-ray based industrial component defect detection method, comprising:

[0130] Processor 201;

[0131] The memory 202 stores computer-readable instructions, and when the computer-readable instructions are executed by the processor 201, the X-ray-based industrial component defect detection method as described in the method embodiment is implemented.

[0132] The X-ray based industrial component defect detection system 20 provided by the present invention can execute the above-mentioned X-ray based industrial component defect detection method and achieve the same or similar technical effects. To avoid repetition, the present invention will not go into details.

[0133] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the X-ray-based industrial component defect detection method as described in the method embodiment is implemented.

[0134] A computer-readable storage medium provided by the present invention can implement the steps and effects of the X-ray-based industrial component defect detection method of the above method embodiment. To avoid repetition, the present invention will not go into details.

Claims

1. An industrial component defect detection method based on X-rays, characterized in that: include: S1: Acquire an X-ray image of the industrial component to be inspected; S2: extracting color features and texture features of the X-ray image; S3: construct a feature vector based on the extracted color features and texture features; S4: Determine the outlier value of each pixel point according to the feature vector of each pixel point; S5: performing defect detection on the industrial component to be detected based on the abnormal value of each pixel point and threshold segmentation to determine the defect position of the industrial component to be detected; The color features include contrast, entropy and grayscale average; the texture features include local kernel regression texture features; The texture feature extraction method specifically includes: The texture features of the X-ray image are extracted by local kernel regression: F i =exp[-(X j -X i ) T C j (X j -X i )] Among them, F i represents the local kernel regression texture feature at the i-th pixel, exp represents the exponential function with a natural constant as the base, X j represents the position coordinates of the jth pixel in the local window, X i Represents the position coordinates of the i-th pixel in the local window, () T represents the matrix transpose operation, C j represents the gradient covariance matrix at the jth pixel in the local window, j∈[1,…,n], n represents the total number of pixels in the local window; The gradient covariance matrix is ​​specifically: in, represents the first-order grayscale derivative along the horizontal direction, represents the first-order grayscale derivative along the vertical direction, Ω i Represents the local window centered at the i-th pixel; Wherein, the S5 is specifically: When the abnormal value of the pixel point is greater than the preset abnormal value and the local variance is less than the preset variance value, it is determined that the position of the current pixel point is a defect position.

2. The X-ray based industrial component defect detection method according to claim 1, characterized in that: The color feature extraction method specifically includes: Obtaining a grayscale histogram of each pixel in the X-ray image under a local window; Determine the occurrence probability of each gray level in the local window; According to the gray level of each pixel in the local window and the probability of occurrence of the gray level, the contrast, entropy value and gray average value of each pixel are calculated: Among them, con i represents the contrast at the i-th pixel, E i Represents the entropy value at the i-th pixel, mean i represents the grayscale average value at the i-th pixel, g represents the grayscale level, P i (g) represents the probability of occurrence of each gray level in the local window centered at the i-th pixel, G represents the gray level value set, N g Indicates the maximum gray level.

3. The X-ray based industrial component defect detection method according to claim 1, characterized in that: The characteristic vector is specifically: I i =[con i ,E i ,mean i ,F i ] Among them, I i Represents the feature vector of the i-th pixel, con i represents the contrast at the i-th pixel, E i Represents the entropy value at the i-th pixel, mean i represents the average grayscale value at the i-th pixel, F i Represents the local kernel regression texture feature at the i-th pixel.

4. The X-ray based industrial component defect detection method according to claim 3, characterized in that: The S4 is specifically: Determine the outlier value of each pixel according to the following formula: Among them, A(X i ) represents the outlier value at the i-th pixel, Ω i represents the local window centered at the i-th pixel, X i Indicates the position coordinates of the i-th pixel, X j represents the position coordinates of the jth pixel in the local window, D represents the difference between pixels, D(X i ,X j ) represents the difference between the i-th pixel and the j-th pixel, w ij Represents the influence coefficient of the j-th pixel on the i-th pixel; The difference is specifically: Among them, con i represents the contrast at the i-th pixel, E i Represents the entropy value at the i-th pixel, mean i represents the average grayscale value at the i-th pixel, F i represents the local kernel regression texture feature at the i-th pixel, con j represents the contrast at the jth pixel, E j Represents the entropy value at the jth pixel, mean j represents the average grayscale value at the jth pixel, F j represents the local kernel regression texture feature at the jth pixel, λ1 represents the weight coefficient of contrast, λ2 represents the weight coefficient of entropy value, λ3 represents the weight coefficient of grayscale average value, and λ4 represents the weight coefficient of local kernel regression texture feature.

5. The X-ray based industrial component defect detection method according to claim 1, characterized in that: The local variance is specifically: Among them, Var i represents the local variance at the i-th pixel, X j Represents the position coordinates of the jth pixel in the local window, A(X i ) represents the outlier value at the jth pixel, Ω i represents the local window centered at the i-th pixel, w ij represents the influence coefficient of the j-th pixel on the i-th pixel, Represents the average value of outliers in the local window centered at the i-th pixel.

6. An industrial component defect detection system based on X-rays, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the X-ray-based industrial component defect detection method according to any one of claims 1 to 5 is implemented.

Images