A defect identification method and system for molybdenum-rhenium alloy pipe based on high-definition microscopic images

Abstract

This invention discloses a method and system for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images. The method includes the following steps: acquiring high-definition images of the molybdenum-rhenium alloy pipe fittings using a high-definition microscope; preprocessing the high-definition images of the molybdenum-rhenium alloy pipe fittings to obtain an image of the molybdenum-rhenium alloy pipe fittings to be detected; performing region growing algorithm processing on the image of the molybdenum-rhenium alloy pipe fittings to be detected to obtain a defect region to be detected; identifying defects in the defect region to be detected to obtain the type of defect; and determining that the molybdenum-rhenium alloy pipe fittings has a defect problem based on the type of defect.

Description

Technical Field

[0001] This invention relates to anomaly detection, and more particularly to a method and system for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images. Background Technology

[0002] Molybdenum-rhenium alloy, as an important high-temperature alloy material, is highly valued for its excellent high-temperature performance, good corrosion resistance, and mechanical strength. Molybdenum-rhenium alloy pipe fittings, as one of its main forms, are often affected by factors such as temperature changes, mechanical stress, and chemical reactions during manufacturing and use, resulting in various types of defects, mainly including porosity and cracks. Porosity defects usually occur during the smelting process when gases fail to escape completely, leading to voids inside or on the surface of the metal. Cracks may occur due to excessive stretching, fatigue, or uneven cooling, causing surface or internal cracking of the metal. These defects not only affect the mechanical properties of molybdenum-rhenium alloy pipe fittings but may also affect their reliability in extreme environments.

[0003] Traditional defect detection methods, such as manual visual inspection or methods based on ordinary image processing, often have a high risk of missed detection and false detection due to insufficient image resolution and the difficulty in identifying small defects, making it difficult to meet the requirements of high precision and high reliability.

[0004] Currently, image segmentation algorithms, as an important defect detection method, have been widely used in various industrial inspections. Region growing algorithms, in particular, can effectively extract defect regions from images by gradually expanding the detection area from a specific seed point. However, surface defects in molybdenum-rhenium alloy pipe fittings have complex morphologies and may resemble the surrounding background. Therefore, improving the segmentation accuracy of region growing algorithms and further identifying the type of defect based on these segmented regions is a key research challenge. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, which solves the above-mentioned technical problems pointed out in the prior art.

[0006] This invention provides a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-resolution microscopic images, comprising the following steps:

[0007] High-definition images of molybdenum-rhenium alloy pipe fittings are acquired using high-definition microscopy equipment; the high-definition images of the molybdenum-rhenium alloy pipe fittings are preprocessed to obtain images of the molybdenum-rhenium to be tested;

[0008] The image of molybdenum and rhenium to be detected is processed by a region growing algorithm to obtain the defect region to be detected; the defect region to be detected is then identified to obtain the type of defect.

[0009] The defects described above indicate that the molybdenum-rhenium alloy pipe fitting has a defect problem.

[0010] Preferredly, as one possible implementation, the types of defects include: porosity type defects and crack type defects.

[0011] Preferredly, as one possible implementation, the image of the molybdenum-rhenium to be detected is processed by a region growing algorithm to obtain the defect region to be detected. The specific operation steps are as follows:

[0012] The Otsu algorithm is used to calculate the gray value of each pixel in the molybdenum-rhenium image to be detected. The pixel frequency is statistically analyzed based on the gray value of the pixel. The pixel frequency is then normalized to obtain the pixel probability.

[0013] A grayscale histogram of the molybdenum-rhenium image to be detected is constructed using pixel frequency.

[0014] The inter-class variance of the separation degree between the foreground and background of the molybdenum-rhenium image to be detected is calculated using the gray-level histogram. The formula for calculating the inter-class variance of the separation degree is as follows:

[0015] ;

[0016] In the formula, t is the threshold, which represents the gray level from 0 to 255;

[0017] This is expressed as the sum of the probabilities of pixels below threshold t;

[0018] It is expressed as the sum of the probabilities of pixels above the threshold t;

[0019] Represented as the average grayscale value below the threshold t;

[0020] The average gray value above the threshold t;

[0021] Iterate through all grayscale values ​​with threshold t ranging from 0 to 255, and calculate the inter-class variance corresponding to the degree of separation. The standard threshold T is obtained, and the calculation formula is:

[0022] ;

[0023] Set the low threshold according to the standard threshold T. With high threshold ;

[0024] According to the low threshold Standard threshold T and high threshold Construct a multi-level threshold sequence;

[0025] Calculate the fuzziness index of each pixel in the image of molybdenum-rhenium to be detected, and construct a fuzzy entropy matrix; extract the feature map of the pixel through the fuzzy entropy matrix; and perform a weighted score on each pixel in the pixel feature to obtain the confidence score of each pixel.

[0026] Seed points are obtained by filtering strong and weak pixels for each pixel in the confidence score by constructing a multi-level threshold sequence;

[0027] The seed points are divided into: strong seed points and weak seed points;

[0028] A region growing algorithm is used to grow the weak seed point in relation to the neighboring weak pixels; the Euclidean distance between the weak seed point and the neighboring weak pixels is calculated.

[0029] A preset Euclidean distance threshold h is set; it is then determined whether the Euclidean distance between the weak seed point and its neighboring weak pixels is less than the preset Euclidean distance threshold h.

[0030] If not, the weak seed point does not meet the growth conditions and stops growing;

[0031] If so, the weak seed point grows on the neighboring weak pixels until the weak seed point no longer meets the growth conditions, then growth stops, and finally the defect region to be detected is obtained, which is the weak defect region to be detected.

[0032] A strong seed point is used to grow neighboring strong pixels; the Euclidean distance between the strong seed point and the neighboring strong pixels is calculated.

[0033] Determine whether the Euclidean distance between the strong seed point and its neighboring strong pixels is less than a preset Euclidean distance threshold h;

[0034] If not, the strong seed point does not meet the growth conditions and stops growing;

[0035] If so, the strong seed point grows on the neighboring weak pixels until the strong seed point no longer meets the growth conditions, at which point growth stops, and finally the defect region to be detected is obtained, which is taken as the strong defect region to be detected.

[0036] Preferredly, as one possible implementation, the defect area to be detected is identified to determine the type of defect. The specific operation steps are as follows:

[0037] The pixels in the weak defect region to be detected are clustered with the same gray value to obtain multiple first pixel clusters; the first pixel clusters are used as suspected pore contours.

[0038] Morphological preprocessing is performed on the weakly detectable defect area to enhance the boundary contour of the suspected pore contour in the weakly detectable defect area.

[0039] The area and perimeter of the suspected pore contour are calculated for the boundary contour of the suspected pore contour.

[0040] The pore roundness of the suspected pore outline is analyzed based on the area and perimeter of the suspected pore outline.

[0041] A preset pore roundness threshold o is used to determine whether the pore roundness of the suspected pore outline is greater than the pore roundness threshold o.

[0042] If not, the suspected pore outline has a regular shape, and it is determined that there is no pore-type defect in the weak defect area to be detected;

[0043] If so, the suspected pore outline is determined to be irregular in shape, and the weakly detectable defect area is determined to have a pore-type defect.

[0044] For the defect area to be detected, the pixel with the largest gray value is selected as the starting pixel. The edge detection algorithm is used to determine whether the gray value of the starting pixel is less than or equal to that of the neighboring pixels.

[0045] If not, the starting pixel stops growing, and the area to be detected has no crack-type defects.

[0046] If so, the starting pixel continues to grow until it forms a crack outline in a region of strong defects to be detected, and the region of strong defects to be detected is determined to have a crack-type defect.

[0047] If the growth of the initial pixel stops during the growth process, it is determined that the area to be detected has no crack-type defects.

[0048] Preferredly, as one possible implementation, the fuzziness index of each pixel in the image of molybdenum-rhenium to be detected is calculated, and a fuzzy entropy matrix is ​​constructed. The feature map of the pixel is extracted using the fuzzy entropy matrix. The specific steps are as follows:

[0049] By setting a multiplication factor, the molybdenum-rhenium image to be detected is decomposed using a Gaussian pyramid according to the multiplication factor to obtain molybdenum-rhenium images at multiple scales.

[0050] The molybdenum-rhenium images at multiple scales are smoothed using windows with Gaussian kernels of different sizes, and then downsampled using bilinear interpolation to obtain downsampled molybdenum-rhenium images at multiple scales.

[0051] Set up a pixel neighborhood window, and calculate the local mean and local standard deviation of the neighboring pixels for each pixel in the molybdenum-rhenium images at multiple scales according to the pixel neighborhood window;

[0052] Calculate the difference between the gray value of each pixel and the local mean; calculate the ambiguity index by normalizing the difference, and generate the ambiguity mapping matrix of the molybdenum-rhenium image at each scale;

[0053] The fuzzy entropy feature map is obtained by using the fuzziness index of each pixel in the fuzzy entropy matrix.

[0054] The variance of the neighboring pixels is further calculated by using the local mean of the neighboring pixels obtained through the neighborhood window of the pixel to obtain the gray-level dispersion of the neighboring pixels; the variance of the neighboring pixels of each pixel is traversed to obtain the gray-level dispersion of the neighboring pixels of each pixel; and the gray-level feature map of each pixel is obtained based on the gray-level dispersion of the neighboring pixels of each pixel.

[0055] Determine the maximum and minimum gray values ​​of the pixels in the neighborhood of each pixel within the neighborhood window; calculate the contrast ratio based on the maximum and minimum gray values; iterate through the neighborhood window of each pixel to calculate the contrast ratio and obtain a contrast feature map.

[0056] Preferredly, as one possible implementation, each pixel in the pixel features is weighted and scored to obtain a confidence score for each pixel; seed points are obtained by filtering each pixel based on its confidence score through a multi-level threshold sequence. The specific steps are as follows:

[0057] The confidence score of each pixel is calculated by weighting the fuzzy entropy feature map, grayscale feature map and contrast feature map of the pixels in the image of molybdenum rhenium to be detected.

[0058] Iterate through each pixel's confidence score to determine if it is greater than the lowest threshold in the multi-level threshold sequence. ;

[0059] If not, then the pixel is a weak pixel and is used as a weak seed point;

[0060] If so, continue to check whether the confidence score of the remaining pixels is greater than the high threshold. ;

[0061] If not, then the pixel is a normal pixel in the image of molybdenum and rhenium to be detected;

[0062] If so, the remaining pixels are used as strong pixels and strong seed points.

[0063] Preferredly, as one possible implementation, the fuzzy entropy feature map, grayscale feature map, and contrast feature map of the pixels in the image of molybdenum and rhenium to be detected are weighted and evaluated to calculate the confidence score of the pixels. The specific operation steps are as follows:

[0064] The fuzzy entropy feature map, grayscale feature map, and contrast feature map are normalized, and then the fuzzy entropy feature map, grayscale feature map, and contrast feature map are merged using a linear weighting method to obtain the fused feature.

[0065] The confidence score of a pixel is obtained by weighting the fused features.

[0066] As a preferred, feasible solution, the specific steps for determining whether the molybdenum-rhenium alloy pipe fitting has a defect based on the aforementioned type of defect are as follows:

[0067] If the image of molybdenum rhenium to be detected has either a porosity defect or a crack defect, then the image of molybdenum rhenium to be detected is determined to have a defect problem.

[0068] Accordingly, the present invention provides a defect identification system for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, comprising: an acquisition module; an identification module; and a conclusion module;

[0069] The acquisition module is used to acquire high-definition images of molybdenum-rhenium alloy pipe fittings using high-definition microscopic equipment; the high-definition images of the molybdenum-rhenium alloy pipe fittings are preprocessed to obtain the image of the molybdenum-rhenium to be detected;

[0070] The identification module is used to perform region growing algorithm processing on the molybdenum-rhenium image to be detected to obtain the defect region to be detected; and to identify defects in the defect region to be detected to obtain the type of defect;

[0071] The conclusion module is used to determine whether the molybdenum-rhenium alloy pipe fitting has a defect problem based on the type of defect.

[0072] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages:

[0073] The present invention provides a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images. In practical applications, it uses high-resolution microscopic equipment to acquire high-resolution images of the pipe fittings, ensuring clear capture of minute defect features (such as pores and cracks), effectively avoiding missed detections or misjudgments due to image blurring, and improving the basic quality of detection. Simultaneously, an improved region growing algorithm is applied, which can accurately extract defect areas in the molybdenum-rhenium alloy pipe fittings, especially defects with intact morphology and strong connectivity (such as pores and cracks). By setting appropriate thresholds and growth conditions, the algorithm exhibits higher sensitivity and accuracy in identifying minute cracks, effectively improving the comprehensiveness and detail of defect detection. By identifying pore-type defects and crack-type defects, the detection method for abnormal defects in molybdenum-rhenium alloy pipe fittings is enhanced. Regardless of the type of defect, the presence of a defect in the pipe fitting can be accurately determined, thereby improving the overall identification quality and the reliability of the judgment.

[0074] In summary, the defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images provided by this invention achieves high-precision and high-efficiency defect detection through high-quality image acquisition, optimized preprocessing steps, accurate region growing algorithm, and comprehensive judgment of multiple types of defects, significantly improving the quality control level of molybdenum-rhenium alloy pipe fittings. Attached Figure Description

[0075] Figure 1 This is an overall flowchart of a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, according to Embodiment 1 of the present invention.

[0076] Figure 2 This is a schematic diagram illustrating the calculation of porosity roundness in a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, according to Embodiment 1 of the present invention.

[0077] Figure 3 This is a schematic diagram of the fish-patterned irregular pores in a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, according to Embodiment 1 of the present invention.

[0078] Figure 4 This is a schematic diagram of weld beads with regular pores in a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, according to Embodiment 1 of the present invention.

[0079] Figure 5 This is a schematic diagram of a defect identification system for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, according to Embodiment 2 of the present invention. Detailed Implementation

[0080] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0081] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0082] Example 1

[0083] See Figure 1 This invention provides a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, comprising the following steps:

[0084] S1: High-definition images of molybdenum-rhenium alloy pipe fittings are acquired using high-definition microscopic equipment; the high-definition images of the molybdenum-rhenium alloy pipe fittings are preprocessed to obtain the images of the molybdenum-rhenium to be tested;

[0085] It should be noted that using high-definition microscopy equipment can acquire high-resolution images of the surface of molybdenum-rhenium alloy pipe fittings, ensuring that minute defect features, such as pores and cracks, can be captured; this is the basis of defect detection, providing sufficiently clear image data to avoid missed or misjudged defects due to blurry images.

[0086] By performing preprocessing steps (such as noise reduction, contrast enhancement, and grayscale conversion) on high-definition images of molybdenum-rhenium alloy pipe fittings, noise in the images can be eliminated and the identifiability of defect features can be enhanced. This is crucial for subsequent region growing algorithms and defect identification, and can improve the accuracy of detection.

[0087] S2: Perform region growing algorithm processing on the molybdenum-rhenium image to be detected to obtain the defect region to be detected; identify defects in the defect region to be detected to obtain the type of defect;

[0088] The types of defects include: porosity defects and crack defects;

[0089] It should be noted that region growing is a common image segmentation method used to gradually add adjacent pixels to the growing region, starting from a seed point. For defect detection in molybdenum-rhenium alloy pipe fittings, the improved region growing algorithm can effectively extract potential defect regions from the image. The technical solution of this application, by setting appropriate thresholds and growth conditions, can identify morphologically complete and connected defect regions (such as pores or cracks), and can also accurately identify minute cracks; it also separates these regions from the background, providing sufficient preparation for subsequent defect type identification.

[0090] S3: The type of defect indicates that the molybdenum-rhenium alloy pipe fitting has a defect problem.

[0091] It should be noted that if the image of the molybdenum rhenium to be detected has either a porosity defect or a crack defect, then the image of the molybdenum rhenium to be detected is determined to have a defect problem.

[0092] Regardless of whether the molybdenum-rhenium image to be tested has a porosity or crack, it can be determined that the image has an abnormal defect. This increases the means of judging abnormal defects in molybdenum-rhenium alloy parts and also improves the identification quality of molybdenum-rhenium alloy parts.

[0093] The improved region growing algorithm is described below:

[0094] Specifically, in step S2, the image of molybdenum and rhenium to be detected is processed by a region growing algorithm to obtain the defect region to be detected; the defect region to be detected is then identified to obtain the type of defect. The specific operation steps are as follows:

[0095] S21: The Otsu algorithm is used to calculate the gray value of each pixel in the image of molybdenum and rhenium to be detected. The pixel frequency is statistically analyzed by the gray value of the pixel. The pixel frequency is normalized to obtain the pixel probability.

[0096] A grayscale histogram of the molybdenum-rhenium image to be detected is constructed using pixel frequency.

[0097] It should be noted that, firstly, the image of molybdenum and rhenium to be detected contains different grayscale regions. When detecting weak or strong boundaries, a single threshold may not be able to effectively distinguish these regions, leading to the loss of important information and affecting the subsequent processing results. Therefore, before performing the region growing algorithm, it is necessary to use the Otsu algorithm to identify different grayscale levels in the image of molybdenum and rhenium to be detected, constructing multi-level thresholds. This addresses the problem that a single threshold cannot effectively identify different grayscale regions, weak or strong boundaries, thus avoiding errors in the defect regions obtained in subsequent processing.

[0098] Therefore, the process begins by iterating through each pixel in the image of the molybdenum-rhenium to be detected. The frequency of each pixel's grayscale value appearing in the image is then calculated. Next, the pixel probability is calculated by comparing the pixel frequency of a particular pixel's grayscale value with the total number of pixels in the image (e.g., if an image contains 1000 pixels, and 200 pixels have a grayscale value of 128, dividing these 200 pixels by the total number of pixels (1000) yields a probability of 0.2 for the pixel with a grayscale value of 128). This process constructs a grayscale histogram, which is a graph describing the number of pixels at each grayscale level (typically 0 to 255) in the image. The purpose of constructing the grayscale histogram is to provide foundational data for subsequent inter-class variance calculations and to help understand the distribution of different grayscale levels in the image.

[0099] S22: The inter-class variance of the separation degree between the foreground and background of the molybdenum-rhenium image to be detected is calculated using the gray-level histogram. The formula for calculating the inter-class variance of the separation degree is as follows:

[0100] ;

[0101] In the formula, t is the threshold, which represents the gray level from 0 to 255;

[0102] It is expressed as the sum of pixel probabilities below the threshold t (i.e., the probability of foreground pixels);

[0103] It is expressed as the sum of the probability of pixels above the threshold t (i.e., the probability of background pixels).

[0104] It is expressed as the average gray value below the threshold t (i.e., the average gray value of the foreground).

[0105] The average gray value above the threshold t (i.e., the average gray value of the background).

[0106] Iterate through all grayscale values ​​with threshold t ranging from 0 to 255, and calculate the inter-class variance corresponding to the degree of separation. The standard threshold T is obtained, and the calculation formula is:

[0107] ;

[0108] It should be noted that in the above formula, the grayscale histogram can reflect the difference between the foreground and background of the molybdenum-rhenium image to be detected through the grayscale level of the pixel (i.e., grayscale from 0 to 255). Therefore, the degree of separation between the foreground and background is calculated by using the inter-class variance. This allows the threshold between the two to be selected as the criterion for selecting the seed pixel in the subsequent process. At the same time, it can also reduce the selection bias caused by the mutual influence between the foreground and background of the molybdenum-rhenium image to be detected.

[0109] Otsu's algorithm is an automatic thresholding method that determines the optimal threshold by maximizing the inter-class variance. It can evaluate the foreground and background discrimination under different thresholds, and thus find the optimal threshold. The threshold that maximizes the inter-class variance is selected as T.

[0110] First, determine the grayscale threshold t. Then, select the foreground and background segmentation point within the grayscale range of 0-255. Calculate the foreground pixel probability (i.e., the foreground pixel probability is the sum of the probabilities of all pixels with grayscale values ​​less than or equal to t). ) and background pixel probability (i.e., the background pixel probability is the sum of the probabilities of all pixels with gray values ​​greater than t: In the formula, Let represent the probability of a pixel with grayscale value i, and N be the total number of pixels in the molybdenum-rhenium image to be detected. This allows us to understand the distribution of different grayscale levels in the image. Then, using the foreground pixel probability and background pixel probability, we calculate the average grayscale value of the foreground and the average grayscale value of the background respectively (average grayscale value of the foreground: ...). Average grayscale value of the background: Then, the inter-class variance is calculated to measure the degree of difference between the foreground and background, and the optimal threshold (i.e., the standard threshold T) is selected.

[0111] S23: Set a low threshold according to the standard threshold T. With high threshold ;

[0112] According to the low threshold Standard threshold T and high threshold Construct a multi-level threshold sequence;

[0113] It should be noted that the low threshold This threshold is used to detect features with low contrast or weak boundaries. It allows more pixels to be allocated to the foreground, helping to capture finer structures and weak edges in the image of molybdenum rhenium to be detected; a higher threshold... Primarily used to ensure strong boundary detection, it is suitable for segmenting high-contrast regions, filtering out noise and ensuring accurate segmentation in high-contrast areas. Low thresholds allow for the capture of weak edge information, while high thresholds ensure the preservation of strong edges. This flexible segmentation strategy enhances the comprehensiveness of edge detection. The constructed multi-level threshold design enables the algorithm to better adapt to various features in images, especially in images of complex molybdenum-rhenium alloy parts, effectively distinguishing regions of different gray levels. Research has found that accurate seed point selection depends on good threshold settings when segmenting and processing defect regions. Therefore, multi-level thresholds ensure that the selected seed points are representative in the feature space, providing a technical foundation for more accurate segmentation of defect morphology and surrounding background.

[0114] The embodiments of this invention suggest that constructing multi-level thresholds can more accurately identify subtle differences exhibited under different lighting conditions or surface characteristics. This is crucial for subsequent region growing algorithms because it provides a more accurate basis for seed point selection, which helps to achieve more refined defect region extraction.

[0115] S24: Calculate the fuzziness index of each pixel in the image of molybdenum-rhenium to be detected, and construct a fuzzy entropy matrix; extract the feature map of the pixel through the fuzzy entropy matrix (i.e., extract the grayscale feature map, fuzzy entropy feature map and contrast feature map); perform a weighted score on each pixel in the feature map of the pixel to obtain the confidence score of each pixel.

[0116] Seed points are obtained by filtering strong and weak pixels for each pixel in the confidence score by constructing a multi-level threshold sequence;

[0117] The seed points are divided into: strong seed points and weak seed points;

[0118] It should be noted that the grayscale deviation of each pixel's grayscale value from the standard threshold T is calculated, and a membership function is constructed based on the grayscale deviation of each pixel's grayscale value. This function can reflect the fuzziness of each pixel, and the fuzzy entropy value of each pixel can be calculated using the membership function. This can indicate the uncertainty of each pixel, making the selection of pixels as seed points clearer and more accurate.

[0119] Furthermore, a fuzzy entropy matrix is ​​constructed, corresponding to the fuzzy entropy value of each pixel; then, feature maps of the pixels are extracted through the fuzzy entropy matrix: grayscale feature map, fuzzy entropy feature map (which can evaluate the information richness of each pixel; measure the complexity and uncertainty of a local region in the image of molybdenum rhenium to be detected, and help identify the details and texture information of the image of molybdenum rhenium to be detected), and contrast feature map; and the confidence score of all pixels is calculated by normalizing and weighting the feature map of each pixel.

[0120] Calculate the low threshold in the multi-level threshold sequence for all pixels. Standard threshold T and high threshold The system performs a judgment and selection process to identify suitable pixels as seed points; if the confidence score of a pixel is greater than a high threshold... If the confidence score is less than the low threshold, then the pixel is considered a strong pixel; If the pixel is selected as a weak pixel, then the strong and weak pixels are selected as strong and weak seed points, respectively, thus obtaining a seed point set. This can adapt to different grayscale distributions and ensure the representativeness of the seed points (i.e., by using a high threshold for screening, noise and misselection can be effectively suppressed, ensuring that the selected seed points have high reliability). At the same time, the seed point selection mechanism ensures that the points are evenly distributed, which is beneficial to subsequent region growth.

[0121] The low threshold in a multi-level threshold sequence is determined by the confidence score of each pixel. and high threshold By filtering out strong and weak pixels and using them as strong and weak seed points, it is also possible to identify crack defects and porosity defects in the defects.

[0122] S25: Use a region growing algorithm to grow the weak seed point and its neighboring weak pixels; calculate the Euclidean distance between the weak seed point and its neighboring weak pixels;

[0123] A preset Euclidean distance threshold h is set; it is then determined whether the Euclidean distance between the weak seed point and its neighboring weak pixels is less than the preset Euclidean distance threshold h.

[0124] If not, the weak seed point does not meet the growth conditions and stops growing;

[0125] If so, the weak seed point grows on the neighboring weak pixels until the weak seed point no longer meets the growth conditions, then growth stops, and finally the defect region to be detected is obtained, which is the weak defect region to be detected.

[0126] A strong seed point is used to grow neighboring strong pixels; the Euclidean distance between the strong seed point and the neighboring strong pixels is calculated.

[0127] Determine whether the Euclidean distance between the strong seed point and its neighboring strong pixels is less than a preset Euclidean distance threshold h;

[0128] If not, the strong seed point does not meet the growth conditions and stops growing;

[0129] If so, the strong seed point grows on the neighboring weak pixels until the strong seed point no longer meets the growth conditions, then growth stops, and finally the defect region to be detected is obtained, which is taken as the strong defect region to be detected.

[0130] It should be noted that stronger pixels among the seed points appear more clearly in the image of the molybdenum-rhenium alloy to be detected, and they magnify smaller defects, making them easier to identify. For example, in molybdenum-rhenium alloy parts, some cracks are quite obvious, due to corrosion or other reasons, and are therefore large and noticeable. However, some cracks are smaller and harder to detect because the molybdenum-rhenium alloy parts are more fragile due to substandard production. Therefore, it is necessary to use strong pixels to highlight these smaller cracks.

[0131] Weak pixels appear weaker in the image of molybdenum-rhenium to be detected, lacking strong grayscale brightness. However, during defect identification, the porosity defects of molybdenum-rhenium alloy parts are relatively prominent and obvious on the surface. Therefore, this invention suggests that using weak pixels to represent the more prominent and easily identifiable porosity defects can better reflect the identification of defects.

[0132] S26: Cluster the pixels in the weak defect region to be detected with the same gray value to obtain multiple first pixel clusters; use the first pixel clusters as suspected pore outlines;

[0133] Morphological preprocessing is performed on the weakly detectable defect area to enhance the boundary contour of the suspected pore contour in the weakly detectable defect area.

[0134] The area and perimeter of the suspected pore contour are calculated for the boundary contour of the suspected pore contour.

[0135] The pore roundness of the suspected pore outline is analyzed based on the area and perimeter of the suspected pore outline.

[0136] A preset pore roundness threshold o is used to determine whether the pore roundness of the suspected pore outline is greater than the pore roundness threshold o.

[0137] If not, the suspected pore outline has a regular shape, and it is determined that there is no pore-type defect in the weak defect area to be detected;

[0138] If so, the suspected pore outline is determined to be irregular in shape, and the weakly detectable defect area is determined to have a pore-type defect.

[0139] It should be noted that clustering pixels with the same grayscale value in the weakly detectable defect area is used to obtain first pixel clusters as potential pores, which can quickly detect the presence of pores. Furthermore, morphological preprocessing is performed on the outlines of potential pores in the weakly detectable defect area to highlight the pore outlines and enhance the boundaries of some blurred pores. This facilitates the calculation of the area and perimeter of the potential pore outline (i.e., the area and perimeter of the potential pore outline are obtained by calculating the radius of the first pixel cluster by the distance from the center pixel to the edge pixels, and the radius is used to obtain the area and perimeter of the potential pore outline). This obtains the basic dimensional features of the potential pore outline and analyzes the pore roundness (i.e., the pore roundness is calculated from the area and perimeter of the potential pore outline, using the formula: Pore roundness = 4π × area / (perimeter)²; for a perfect circle, such as...). Figure 2 As shown, this value equals 1; the more irregular the shape, the closer the value is to 0; the calculation is simple and suitable for rapid evaluation), assessing the regularity of porosity. More regular porosity often indicates more serious defects (i.e., the welding of molybdenum-rhenium alloy parts often has a fish-scale pattern, such as...). Figure 3 As shown, if pores appear, they will be circular in shape. Therefore, the more regular the pore shape, the more severe the condition. Figure 4 (as shown)

[0140] S27: Select the pixel with the largest gray value in the defect area to be detected as the starting pixel, and use the edge detection algorithm to determine whether the gray value of the starting pixel is less than or equal to the neighboring pixels.

[0141] If not, the starting pixel stops growing, and the area to be detected has no crack-type defects.

[0142] If so, the starting pixel continues to grow until it forms a crack outline in a region of strong defects to be detected, and the region of strong defects to be detected is determined to have a crack-type defect.

[0143] If the growth of the initial pixel stops during the growth process, it is determined that the area to be detected has no crack-type defects (that is, when the initial pixel grows to half its length, if the gray value of the intermediate pixel is greater than the gray value of the previously grown pixel, then growth stops at this point and the crack outline of the area to be detected cannot be formed, so it is determined that the area to be detected has no crack-type defects).

[0144] It should be noted that the edge detection algorithm is used to form an outline of the area to be detected. If a complete area to be detected is formed, it is equivalent to a verification of the overlap of the crack in the area to be detected. When the two overlap (that is, the area to be detected is wrapped, indicating that a double verification has been performed), it means that there is a crack in the area to be detected. If the crack outline only grows to half, it means that the two do not overlap, and there is no crack in the area to be detected.

[0145] This application's embodiments reveal that traditional region growing algorithms rely on the selection of initial seed points. Inappropriate seed point selection can lead to misjudgments or missed detections. Particularly in complex molybdenum-rhenium alloy pipe images, the grayscale difference between the defect area and the background can be very small, making seed point selection even more difficult. Therefore, this application's embodiments propose processing steps to improve the accuracy of seed point selection and the ability to distinguish detailed defect features, as detailed in subsequent steps S241-S244.

[0146] Specifically, in step S24, the fuzziness index of each pixel in the image of molybdenum-rhenium to be detected is calculated to construct a fuzzy entropy matrix; feature maps of the pixels are extracted using the fuzzy entropy matrix; each pixel in the pixel feature is weighted and scored to obtain a confidence score for each pixel; seed points are obtained by constructing a multi-level threshold sequence to filter each pixel with a confidence score. The specific operation steps are as follows:

[0147] S241: Set the multiplication factor, and use the Gaussian pyramid to decompose the molybdenum-rhenium image to be detected according to the multiplication factor to obtain molybdenum-rhenium images at multiple scales;

[0148] The molybdenum-rhenium images at multiple scales are smoothed using windows with Gaussian kernels of different sizes, and then downsampled using bilinear interpolation to obtain downsampled molybdenum-rhenium images at multiple scales.

[0149] Set up a pixel neighborhood window, and calculate the local mean and local standard deviation of the neighboring pixels for each pixel in the multi-scale molybdenum-rhenium images (i.e., downsampled multi-scale molybdenum-rhenium images);

[0150] Calculate the difference between the gray value of each pixel and the local mean; calculate the fuzziness index by normalizing the difference, and generate the fuzziness mapping matrix (i.e., fuzzy entropy matrix) of the molybdenum-rhenium image at each scale.

[0151] It should be noted that a multiplication factor is set (i.e., constructing four scale levels [1,2,4,8] to cover multiple receptive fields, ensuring the capture of image features at different scales). A Gaussian pyramid is used to decompose the molybdenum-rhenium image under test according to the multiplication factor, resulting in multiple scale images of molybdenum-rhenium, improving multi-scale analysis and enhancing feature description. Each scale uses a Gaussian kernel of different sizes for smoothing to avoid over-smoothing and information loss. Downsampling is performed at each scale level to generate image representations at different resolutions (i.e., downsampled molybdenum-rhenium images at multiple scales). A pixel neighborhood window is set (i.e., the window can be set to 3×3, and local regional feature information is calculated for pixels within this window). The local mean μ and standard deviation σ within the 3×3 window are calculated, and then... The difference between the pixel's grayscale value and the local mean is calculated (i.e., the absolute value of the grayscale value of the current pixel minus the local mean is divided by the local standard deviation; the purpose of dividing by the standard deviation is to normalize the difference value). The difference is then converted into a fuzziness metric (i.e., a fuzziness index or fuzziness value). Fuzziness indicates the degree to which a pixel belongs to a certain region. A fuzziness mapping matrix is ​​generated at each scale, and the size of the matrix varies with the scale. Assuming a 100×100 image: the matrix sizes at different scales are: Scale 1: 100×100, Scale 2: 50×50, Scale 4: 25×25, Scale 8: 13×13. The numerical meaning of each position is: 0.9 represents a highly uncertain region, 0.5 represents a moderately uncertain region, and 0.1 represents a relatively certain region.

[0152] S242: Obtain the fuzzy entropy feature map by using the fuzziness index of each pixel in the fuzzy entropy matrix;

[0153] The variance of the neighboring pixels is further calculated by using the local mean of the neighboring pixels obtained through the neighborhood window of the pixel to obtain the gray-level dispersion of the neighboring pixels; the variance of the neighboring pixels of each pixel is traversed to obtain the gray-level dispersion of the neighboring pixels of each pixel; and the gray-level feature map of each pixel is obtained based on the gray-level dispersion of the neighboring pixels of each pixel.

[0154] Determine the maximum and minimum gray values ​​of the pixels in the neighborhood of each pixel within the neighborhood window; calculate the difference contrast based on the maximum and minimum gray values; iterate through and calculate the difference contrast within the neighborhood window of each pixel to obtain a contrast feature map.

[0155] It should be noted that the fuzzy entropy matrix is ​​used to extract the fuzziness index of each pixel. First, since the fuzziness index of each pixel has already been calculated, it is directly used as the extracted fuzzy entropy feature map. The fuzzy entropy feature map can evaluate the information richness of each pixel. Pixels with high fuzzy entropy values ​​are usually located in complex regions and can provide more information, which helps to improve the representativeness of the seed point. Thus, a fuzzy entropy feature map is constructed to measure the complexity and uncertainty of a local region in the molybdenum-rhenium image to be detected, and to help identify the details and texture information of the molybdenum-rhenium image to be detected.

[0156] By using a pixel neighborhood window to identify local grayscale feature maps in the image of molybdenum and rhenium to be detected, a grayscale feature map is constructed, providing information on the brightness distribution in the image, which is used to represent the basic visual features of the image of molybdenum and rhenium to be detected.

[0157] By calculating the grayscale contrast between pixels, the overall visual effect of the molybdenum-rhenium image to be detected can be reflected, making the image more vivid and the details more prominent, thus making it more visually appealing; the contrast feature map measures the intensity of brightness changes in local areas, reflecting the structural and edge information in the image;

[0158] S243: The fuzzy entropy feature map, grayscale feature map and contrast feature map of the pixels in the image of molybdenum rhenium to be detected are weighted to evaluate the pixels and calculate the confidence score of the pixels.

[0159] It should be noted that, based on the features extracted from pixels, the fuzzy entropy feature map improves detail recognition, the grayscale feature map provides basic detection capabilities, and the contrast feature map enhances boundary detection. Therefore, by normalizing the fuzzy entropy feature map, the grayscale feature map, and the contrast feature map, the three can be weighted in one dimension. The recognition capabilities of the three are used to evaluate the pixels, thereby determining the position or recognition information (i.e., defect information) of each pixel in the image of molybdenum and rhenium to be detected, and better performing fusion calculation of pixel confidence scores.

[0160] S244: Iterate through the confidence score of each pixel to determine whether it is greater than the low threshold in the multi-level threshold sequence. ;

[0161] If not, then the pixel is a weak pixel and is used as a weak seed point;

[0162] If so, continue to check whether the confidence score of the remaining pixels is greater than the high threshold. ;

[0163] If not, then the pixel is a normal pixel in the image of molybdenum and rhenium to be detected;

[0164] If so, then the remaining pixels are used as strong pixels and strong seed points;

[0165] It should be noted that the selection is performed layer by layer through a multi-level threshold sequence. First, the lowest threshold is used. The confidence scores of pixels are evaluated, and pixels with lower confidence scores are selected as weak pixels. These weak pixels are used as weak seed points to generate the defect region to be detected for porosity defects, thus demonstrating the identification of porosity defects. The remaining pixels are then evaluated to see if their scores exceed a high threshold. Strong pixels are selected as strong seed points to identify the defect area to be detected, demonstrating the identification of crack defects; while the confidence score is at a low threshold. With high threshold Pixels between these points are considered normal pixels and may not have any defect issues.

[0166] Further research revealed that, for example, fuzzy entropy feature maps may be more suitable for identifying subtle changes in complex regions, while contrast feature maps are better suited for capturing edge information. How to rationally combine these features to allow them to leverage their respective strengths in the final confidence score is a problem that needs to be solved. Since the surface of molybdenum-rhenium alloy pipe fittings may contain various complex textures and minute defects, this embodiment of the invention proposes an expanded information processing procedure in steps S2431-S242 based on step S243:

[0167] Specifically, in step S243, the fuzzy entropy feature map, grayscale feature map, and contrast feature map of the pixels in the image of molybdenum rhenium to be detected are weighted and evaluated to calculate the confidence score of the pixels. The specific operation steps are as follows:

[0168] S2431: Normalize the fuzzy entropy feature map, grayscale feature map, and contrast feature map, and merge the fuzzy entropy feature map, grayscale feature map, and contrast feature map using a linear weighting method to obtain fused features;

[0169] It should be noted that when extracting the fuzzy entropy feature map, grayscale feature map, and contrast feature map for each pixel, since the value ranges of different features may be different, we need to normalize each feature to make their value ranges consistent, which will facilitate subsequent weighted calculations.

[0170] The range of grayscale values ​​is usually [0, 255]. To unify the range of grayscale feature maps, they are normalized to [0, 1]. The value of fuzzy entropy is usually within a certain range, assuming its range is [0, max(H)]. It can be normalized using the following formula: This unifies the range of fuzzy entropy feature map values ​​to [0,1]; the range of local contrast may be large, which can be normalized using a similar method: This unifies the range of the contrast feature map to [0,1]. Once the feature values ​​are standardized, these features can be merged using a linear weighting method to obtain the fused features.

[0171] S2432: The confidence score of the pixel is obtained by weighted evaluation using the fused features;

[0172] It should be noted that the fused features are weighted, assuming that the fuzzy entropy feature map, grayscale feature map and contrast feature map are assigned weights wH, wG and wC, satisfying the condition: wH+wG+wC=1, and the confidence score is calculated for each pixel in the molybdenum rhenium image to be detected.

[0173] To improve the rationality of feature fusion and overcome the potential difference in importance between different features for defect detection, step S2431 introduces a linear weighting method. This method assigns different weights to each feature based on actual needs. For example, crack defects typically exhibit high-contrast, elongated structures, thus the contrast feature map can be assigned a higher weight. Conversely, for porosity defects, the fuzzy entropy feature map may be more important, so its weight can be appropriately increased. This flexible weighting method allows the algorithm to dynamically adjust the importance of features based on different types of defects, thereby improving the rationality of feature fusion. Normalization ensures that the value range of all features is unified to [0, 1], guaranteeing that each feature contributes the same weight in subsequent weighted calculations. This prevents a single feature from dominating the entire evaluation process due to an excessively large value range, ensuring that all features participate fairly in the final confidence score calculation.

[0174] Furthermore, by employing multi-feature fusion, combining features such as fuzzy entropy, grayscale, and local contrast, the characteristics of each pixel can be described more comprehensively. The introduction of normalization and weighted evaluation enables the algorithm to maintain stable performance under different conditions, enhancing its adaptability to complex images. At the same time, reasonable weight allocation improves the algorithm's robustness and reduces the impact of noise and other interference factors.

[0175] In summary, the technical solution of this invention, in step S21, uses the Otsu algorithm to calculate the grayscale histogram and select the optimal threshold T to better separate the foreground (i.e., potential defects) and the background (i.e., normal pipe surfaces), thereby avoiding the loss of defect information caused by a single threshold failing to effectively distinguish regions of different grayscale levels. By constructing multi-level thresholds, subtle differences exhibited under different lighting conditions or surface characteristics can be identified more accurately. This is crucial for subsequent region growing algorithms because it provides a more accurate basis for seed point selection, facilitating more refined defect region extraction.

[0176] Furthermore, by introducing the concepts of fuzzy entropy and confidence score, the understanding of pixel features is further enhanced, allowing the system to perform weighted scoring based on the uncertainty of each pixel, thereby selecting suitable strong and weak seed points. This method not only improves the ability to identify different types of defects such as cracks and porosity, but also ensures high detection accuracy even under low contrast or high noise conditions. Finally, by analyzing the roundness of suspected porosity contours and confirming crack contours using edge detection algorithms, effective classification of the two main types of defects is achieved, increasing the reliability and accuracy of quality assessment for molybdenum-rhenium alloy pipe fittings.

[0177] Example 2

[0178] like Figure 5 As shown, the present invention also provides a defect identification system for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, comprising: an acquisition module 10; an identification module 20; and a conclusion module 30.

[0179] The acquisition module 10 is used to acquire high-definition images of molybdenum-rhenium alloy pipe fittings using high-definition microscope equipment; and to preprocess the high-definition images of molybdenum-rhenium alloy pipe fittings to obtain the image of molybdenum-rhenium to be detected.

[0180] The identification module 20 is used to perform region growing algorithm processing on the molybdenum-rhenium image to be detected to obtain the defect region to be detected; and to identify defects in the defect region to be detected to obtain the type of defect;

[0181] The conclusion module 30 is used to determine whether the molybdenum-rhenium alloy pipe fitting has a defect problem based on the type of defect.

[0182] The present invention provides a method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images. Through high-quality image acquisition, optimized preprocessing steps, accurate region growing algorithm, and comprehensive judgment of multiple types of defects, it achieves high-precision and high-efficiency defect detection, and significantly improves the quality control level of molybdenum-rhenium alloy pipe fittings.

[0183] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for defect identification of molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, characterized in that, The following steps are included: High-definition images of molybdenum-rhenium alloy pipe fittings are acquired using high-definition microscopy equipment; the high-definition images of the molybdenum-rhenium alloy pipe fittings are preprocessed to obtain images of the molybdenum-rhenium to be tested; The image of molybdenum and rhenium to be detected is processed by a region growing algorithm to obtain the defect region to be detected; the defect region to be detected is then identified to obtain the type of defect. The type of defect indicates that the molybdenum-rhenium alloy pipe fitting has a defect problem; The types of defects include: porosity defects and crack defects; The image of the molybdenum-rhenium to be detected is processed by a region growing algorithm to obtain the defect region to be detected. The specific operation steps are as follows: The Otsu algorithm is used to calculate the gray value of each pixel in the molybdenum-rhenium image to be detected. The pixel frequency is statistically analyzed based on the gray value of the pixel. The pixel frequency is then normalized to obtain the pixel probability. A grayscale histogram of the molybdenum-rhenium image to be detected is constructed using pixel frequency. The inter-class variance of the separation degree between the foreground and background of the molybdenum-rhenium image to be detected is calculated using the gray-level histogram. The formula for calculating the inter-class variance of the separation degree is as follows: ； In the formula, t is the threshold value, which represents the gray level from 0 to 255; This is expressed as the sum of the probabilities of pixels below threshold t; It is expressed as the sum of the probabilities of pixels above the threshold t; Represented as the average grayscale value below the threshold t; The average gray value above the threshold t; Iterate through all grayscale values ​​with threshold t ranging from 0 to 255, and calculate the inter-class variance corresponding to the degree of separation. The standard threshold T is obtained, and the calculation formula is: ； Set the low threshold according to the standard threshold T. With high threshold ; According to the low threshold Standard threshold T and high threshold Construct a multi-level threshold sequence; Calculate the fuzziness index of each pixel in the image of molybdenum-rhenium to be detected, and construct a fuzzy entropy matrix; extract the feature map of the pixel through the fuzzy entropy matrix; and perform a weighted score on each pixel in the pixel feature to obtain the confidence score of each pixel. Seed points are obtained by filtering strong and weak pixels for each pixel in the confidence score by constructing a multi-level threshold sequence; The seed points are divided into: strong seed points and weak seed points; A region growing algorithm is used to grow the weak seed point in relation to its neighboring weak pixels; the Euclidean distance between the weak seed point and its neighboring weak pixels is calculated. A preset Euclidean distance threshold h is set; it is then determined whether the Euclidean distance between the weak seed point and its neighboring weak pixels is less than the preset Euclidean distance threshold h. If not, the weak seed point does not meet the growth conditions and stops growing; If so, the weak seed point grows on the neighboring weak pixels until the weak seed point no longer meets the growth conditions, then growth stops, and finally the defect region to be detected is obtained, which is the weak defect region to be detected. A strong seed point is used to grow neighboring strong pixels; the Euclidean distance between the strong seed point and the neighboring strong pixels is calculated. Determine whether the Euclidean distance between the strong seed point and the neighboring strong pixels is less than a preset Euclidean distance threshold h; If not, the strong seed point does not meet the growth conditions and stops growing; If so, the strong seed point grows on the neighboring weak pixels until the strong seed point no longer meets the growth conditions, at which point growth stops, and finally the defect region to be detected is obtained, which is taken as the strong defect region to be detected.

2. The defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images according to claim 1, characterized in that, The defects in the area to be detected are identified to determine the type of defect. The specific operation steps are as follows: The pixels in the weak defect region to be detected are clustered with the same gray value to obtain multiple first pixel clusters; the first pixel clusters are used as suspected pore contours. Morphological preprocessing is performed on the weakly detectable defect region to enhance the boundary contour of the suspected pore contour in the weakly detectable defect region. The area and perimeter of the suspected pore contour are calculated for the boundary contour of the suspected pore contour. The pore roundness of the suspected pore outline is analyzed based on the area and perimeter of the suspected pore outline. A preset pore roundness threshold o is used to determine whether the pore roundness of the suspected pore outline is greater than the pore roundness threshold o. If not, the suspected pore outline has a regular shape, and it is determined that there is no pore-type defect in the weak defect area to be detected; If so, the suspected pore outline is determined to be irregular in shape, and the weakly detectable defect area is determined to have a pore-type defect. For the defect area to be detected, the pixel with the largest gray value is selected as the starting pixel. The edge detection algorithm is used to determine whether the gray value of the starting pixel is less than or equal to that of the neighboring pixels. If not, the starting pixel stops growing, and the area to be detected has no crack-type defects. If so, the starting pixel continues to grow until it forms a crack outline in a region of strong defects to be detected, and the region of strong defects to be detected is determined to have a crack-type defect. If the growth of the initial pixel stops during the growth process, it is determined that the area to be detected has no crack-type defects.

3. The defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images according to claim 2, characterized in that, The blur index of each pixel in the image of molybdenum-rhenium to be detected is calculated, and a fuzzy entropy matrix is ​​constructed. The feature map of the pixel is then extracted using the fuzzy entropy matrix. The specific steps are as follows: By setting a multiplication factor, the molybdenum-rhenium image to be detected is decomposed using a Gaussian pyramid according to the multiplication factor to obtain molybdenum-rhenium images at multiple scales. The molybdenum-rhenium images at multiple scales are smoothed using windows with Gaussian kernels of different sizes, and then downsampled using bilinear interpolation to obtain downsampled molybdenum-rhenium images at multiple scales. Set up a pixel neighborhood window, and calculate the local mean and local standard deviation of the neighboring pixels for each pixel in the molybdenum-rhenium images at multiple scales according to the pixel neighborhood window; Calculate the difference between the gray value of each pixel and the local mean; calculate the ambiguity index by normalizing the difference, and generate the ambiguity mapping matrix of the molybdenum-rhenium image at each scale; The fuzzy entropy feature map is obtained by using the fuzziness index of each pixel in the fuzzy entropy matrix. The variance of the neighboring pixels is further calculated by using the local mean of the neighboring pixels obtained through the neighborhood window of the pixel to obtain the gray-level dispersion of the neighboring pixels; the variance of the neighboring pixels of each pixel is traversed to obtain the gray-level dispersion of the neighboring pixels of each pixel; and the gray-level feature map of each pixel is obtained based on the gray-level dispersion of the neighboring pixels of each pixel. Determine the maximum and minimum grayscale values ​​of a pixel in the neighborhood of a pixel within the neighborhood window; The difference contrast is calculated based on the maximum and minimum pixel gray values; the difference contrast within the neighborhood window of each pixel is calculated to obtain the contrast feature map.

4. The defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images according to claim 3, characterized in that, Each pixel in the pixel feature is weighted and scored to obtain a confidence score for each pixel; seed points are obtained by filtering each pixel based on its confidence score through a multi-level threshold sequence. The specific steps are as follows: The confidence score of each pixel is calculated by weighting the fuzzy entropy feature map, grayscale feature map and contrast feature map of the pixels in the image of molybdenum rhenium to be detected. Iterate through each pixel's confidence score to determine if it is greater than the lowest threshold in the multi-level threshold sequence. ; If not, then the pixel is a weak pixel and is used as a weak seed point; If so, continue to check whether the confidence score of the remaining pixels is greater than the high threshold. ; If not, then the pixel is a normal pixel in the image of molybdenum and rhenium to be detected; If so, the remaining pixels are used as strong pixels and strong seed points.

5. The defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images according to claim 4, characterized in that, The confidence score of each pixel in the image of molybdenum-rhenium to be detected is calculated by weighting the fuzzy entropy feature map, grayscale feature map, and contrast feature map of the pixels, and the specific operation steps are as follows: The fuzzy entropy feature map, grayscale feature map, and contrast feature map are normalized, and then the fuzzy entropy feature map, grayscale feature map, and contrast feature map are merged using a linear weighting method to obtain the fused feature. The confidence score of a pixel is obtained by weighting the fused features.

6. The defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images according to claim 5, characterized in that, The specific steps for determining whether the molybdenum-rhenium alloy pipe fitting has a defect based on the aforementioned type of defect are as follows: If the image of molybdenum rhenium to be detected has either a porosity defect or a crack defect, then the image of molybdenum rhenium to be detected is determined to have a defect problem.

7. A defect identification system for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images, used to implement the defect identification method for molybdenum-rhenium alloy pipe fittings based on high-definition microscopic images according to any one of claims 1-6, characterized in that, include: Data acquisition module; Recognition module; Conclusion module; The acquisition module is used to acquire high-definition images of molybdenum-rhenium alloy pipe fittings using high-definition microscopic equipment; High-resolution images of molybdenum-rhenium alloy pipe fittings are preprocessed to obtain images of the molybdenum-rhenium to be tested; The identification module is used to perform region growing algorithm processing on the molybdenum-rhenium image to be detected to obtain the defect region to be detected; and to identify defects in the defect region to be detected to obtain the type of defect; The conclusion module is used to determine whether the molybdenum-rhenium alloy pipe fitting has a defect problem based on the type of defect.

Images