Intelligent detection method for high-precision welding of aircraft aluminum condenser

Abstract

The invention relates to the technical field of image processing, and discloses an intelligent detection method for high-precision welding of an aircraft aluminum condenser, which comprises the following steps: acquiring an image of a to-be-detected welding area and preprocessing to obtain a welding area sub-image; calculating a gradient magnitude graph and a local formal dimension graph; the product of the pixel gradient value and the corresponding pixel value is used as a composite density function, and the composite density weighted Zernike moment of each order is calculated; the composite density weighted Zernike moment of each order is adjusted, and a direction enhanced composite Zernike feature is generated; carrying out convolution on the sub-image of the welding area to generate a group of Gabor energy feature maps, and calculating multi-scale Gabor-Hu texture features; and fusing the Zernike features with the multi-scale Gabor-Hu texture features to form a high-dimensional composite feature vector, and inputting the high-dimensional composite feature vector into a pre-trained classification model to output a quality detection category of the welding area. According to the method, the gradient amplitude is combined with the local shape dimension, so that the Zernike moment can simultaneously describe the macroscopic contour and the microscopic details of the welding seam, and the accuracy of welding defect detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to an intelligent detection method for high-precision welding of aircraft aluminum condensers. Background Art

[0002] Aircraft aluminum condensers are heat exchange components in the aircraft's environmental control system, and their quality is related to the stable operation of the entire system and even flight safety. Welding is a process in the condenser manufacturing process. Due to the characteristics of aluminum alloy materials and the complexity of the welding process, the welding area is very prone to various defects such as pores, cracks, lack of fusion, weld bumps and inconsistent surface morphology. These defects will seriously affect the sealing, pressure resistance and heat exchange efficiency of the condenser, posing a major safety hazard. The detection methods in the existing technology mainly rely on X-ray, ultrasonic flaw detection and other means. Although detection methods such as X-ray and ultrasonic flaw detection have high accuracy, they have problems such as expensive equipment, complex operation, long detection cycle, and insensitivity to certain types of surface or near-surface defects.

[0003] Intelligent inspection methods based on image processing offer the advantages of non-contact, high efficiency, and high objectivity in welding quality inspection. They can extract stable and highly discriminative features from weld images with complex textures and uneven reflections. In existing machine vision-based welding inspection technologies, feature extraction is a key factor in determining inspection performance. Existing machine vision-based welding inspection technologies have explored a variety of feature descriptors, such as texture features like gray-level co-occurrence matrices and local binary patterns, as well as shape features like Hough transforms and edge profiles. However, a single type of feature often fails to fully describe the diversity and complexity of welding defects.

[0004] Moment functions, as a classic regional shape descriptor, are used in image analysis due to their excellent geometric invariance. Zernike moments, due to their orthogonality and rotation invariance, excel in target recognition. However, standard Zernike moments treat the image as a uniform density distribution during calculation, ignoring important gradient and texture details within the weld region and making it difficult to effectively characterize the anisotropic characteristics of the weld along the welding direction. Although Hu invariant moments are invariant to translation, scale, and rotation, as global feature descriptors, they are limited in their ability to describe local information rich in texture details on the uneven surface of the weld. Furthermore, Gabor filters are very effective in texture analysis, but effectively integrating Gabor filters with shape features to enhance the comprehensive ability to detect complex weld defects remains a challenge. Therefore, coupling global features that reflect weld contour morphology with local features that characterize surface texture and directionality is a technical challenge that needs to be urgently addressed in the field of intelligent detection of aircraft condenser welding quality. Summary of the Invention

[0005] The present invention provides an intelligent detection method for high-precision welding of aircraft aluminum condensers to solve the problem in the above-mentioned prior art that it is difficult to couple the global features reflecting the weld profile morphology with the local features that can depict the surface texture and directionality.

[0006] The intelligent detection method for high-precision welding of aircraft aluminum condensers of the present invention comprises the following steps: The invention obtains an image of the welding area of ​​the condenser to be inspected, pre-processes the image, and obtains a normalized welding area sub-image; calculates the gradient amplitude map of the welding area sub-image, and uses the box counting method to calculate the local fractal dimension map of the pixels in the gradient amplitude map; takes the geometric center of the welding area sub-image as the origin, multiplies the pixel gradient value in the gradient amplitude map by the corresponding pixel value in the local fractal dimension map as the composite density function, and calculates the composite density weighted Zernike moments of each order; sets a non-uniform angular frequency weight matrix that matches the main welding direction, and uses the non-uniform angular frequency weight matrix to adjust the composite density weighted Zernike moments of each order to enhance the compatibility with the main welding direction. The method is simple, quick, and reliable. The method is convenient to use. The method is simple, quick, and reliable. The method is convenient to use. The method is convenient to use. The method is convenient to use. The method is convenient to use. The method is convenient to use. The method is convenient to use. The method is convenient to use. The method is convenient to use.

[0007] Preferably, the preprocessing of the image to obtain a normalized welding area sub-image includes: converting the acquired image into a grayscale image; using a The grayscale image is filtered using a windowed median filter; the Canny operator is used to perform edge detection on the filtered grayscale image to identify the weld contour; a 256×256 pixel weld area sub-image is cropped from the grayscale image based on the weld contour, and the pixel values ​​of the weld area sub-image are linearly normalized to the interval [0, 1].

[0008] Preferably, the method of calculating the local fractal dimension map of pixels in the gradient amplitude map by using the box counting method includes: for each pixel in the gradient amplitude map, taking a pixel as the center An analysis window of pixels; within the analysis window, set a set of boxes of size S {2, 4, 8, 16}; for each box size S, calculate the minimum number of boxes required to cover all non-zero gradient pixels within the analysis window ; In is the horizontal axis, In the double logarithmic coordinate system with y as the ordinate, the least squares linear fitting is performed on the data points, and the absolute value of the slope of the obtained straight line is used as the local fractal dimension of the central pixel.

[0009] Preferably, the method of adjusting the composite density weighted Zernike moments of each order by using the non-uniform angular frequency weight matrix includes: performing Radon transformation on the gradient amplitude map to determine the main welding direction. ; Weighted Zernike moments of composite density of each order Multiply by the twiddle factor To achieve rotation normalization; a non-uniform angular frequency weight matrix is ​​used to weight the Zernike moments after rotation normalization.

[0010] Preferably, the Gabor filter group consists of 5 center frequencies and 8 direction angles, wherein the center frequencies are {1 / 4, 1 / 8, 1 / 16, 1 / 32, 1 / 64}, the unit of the center frequency is cycle / pixel, and the direction angles are {0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, 157.5°}.

[0011] Preferably, the constructing of multi-scale Gabor-Hu texture features includes: calculating the first to seventh order Hu invariant moments for each image in the set of Gabor energy feature maps; and cascading all calculated Hu invariant moments into a one-dimensional feature vector in a predetermined order.

[0012] Preferably, the classification model is a random forest classification model.

[0013] Preferably, the classification model outputs the quality inspection category of the welding area, including: the random forest classification model has 500 decision trees, and the maximum depth of each decision tree is 25, and the node splitting criterion is Gini impurity; the quality inspection category output by the random forest classification model is "qualified", "surface porosity", "incomplete penetration" or "bite".

[0014] Preferably, the acquiring of the image of the welding area of ​​the condenser to be inspected includes: using a camera in conjunction with a coaxial light source to capture an RGB image of the welding area of ​​the condenser to be inspected.

[0015] Preferably, the image is preprocessed to obtain a normalized welding area sub-image, including: converting the RGB image into a grayscale image, smoothing and denoising using a Gaussian filter, cropping the welding area in the grayscale image through a bilinear interpolation algorithm and uniformly scaling it to a size of 256×256 pixels to obtain a welding area sub-image, and linearly normalizing the pixel values ​​of the welding area sub-image to the interval [0, 1].

[0016] The beneficial effects of the present invention are as follows: by constructing a composite density-weighted Zernike moment, the present invention combines the gradient amplitude reflecting the weld edge information with the local fractal dimension characterizing the complexity of its surface texture, so that the improved Zernike moment can simultaneously describe the macroscopic outline and microscopic details of the weld; in addition, by introducing a non-uniform angular frequency weight matrix that matches the main welding direction, the Zernike moment is directionally enhanced and adjusted, thereby improving the sensitivity of the feature to directional defects such as cracks and lack of fusion. A multi-scale Gabor-Hu texture feature is also constructed, and by calculating the Hu invariant moment of a set of Gabor energy maps, detailed features such as the weld surface roughness and weld wave texture are obtained. The directionally enhanced shape feature is integrated with the multi-scale invariant texture feature to form a composite feature with strong information complementary discriminant power, which reduces the impact of the limitations of a single feature and improves the accuracy of complex welding defect detection in aircraft aluminum condensers. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic flow chart of an intelligent detection method for high-precision welding of aircraft aluminum condensers provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0019] like Figure 1 As shown, the intelligent detection method for high-precision welding of aircraft aluminum condensers provided by the embodiment of the present invention specifically includes the following steps: S1, obtaining an image of the welding area of ​​the condenser to be inspected, preprocessing the image to obtain a normalized welding area sub-image; calculating the gradient amplitude map of the welding area sub-image, and using the box counting method to calculate the local fractal dimension map of the pixels in the gradient amplitude map.

[0020] Specifically, a camera is used in conjunction with a coaxial light source to collect an RGB image of the welding area of ​​the condenser to be inspected. The RGB image is converted into a grayscale image, and a Gaussian filter is used for smoothing and denoising. The welding area in the grayscale image is cropped and uniformly scaled to 256×256 pixels using a bilinear interpolation algorithm to obtain a welding area sub-image. The pixel values ​​of the welding area sub-image are linearly normalized to the interval [0, 1].

[0021] Calculate the gradient of the normalized welding area sub-image in the x and y directions respectively, and synthesize the gradient magnitude map. Take each pixel in the gradient magnitude map as the center, and calculate the gradient of the pixel Within a neighborhood, use boxes of different sizes to cover the neighborhood and count the number of boxes required, e.g. 、 、 、 By performing a linear regression fit on the logarithm of the box size and the logarithm of the number of boxes, the slope of the resulting line is the local fractal dimension of the central pixel, and the local fractal dimension map is obtained.

[0022] S2, taking the geometric center of the welding area sub-image as the origin, taking the product of the pixel gradient value in the gradient amplitude map and the corresponding pixel value in the local fractal dimension map as the composite density function, and calculating the composite density weighted Zernike moments of each order.

[0023] Specifically, the gradient magnitude map is multiplied point-by-point with the corresponding pixel values ​​in the local fractal dimension map to produce a composite density function image. The coordinates of the weld region subimage are mapped to the unit disk. Based on the definition of Zernike moments, the composite density function replaces the image grayscale function in the traditional definition and calculates all Zernike moments from order 0 to the preset highest order.

[0024] S3, set a non-uniform angular frequency weight matrix that matches the main welding direction, use the non-uniform angular frequency weight matrix to adjust the composite density weighted Zernike moments of each order, enhance the angular frequency components associated with the main welding direction, and generate direction-enhanced composite Zernike features; use a group of Gabor filter groups with different center frequencies and direction angles to convolve the welding area sub-image to generate a group of Gabor energy feature maps, calculate seven Hu invariant moments for each Gabor energy feature map, and cascade all Hu invariant moments to form a multi-scale Gabor-Hu texture feature.

[0025] For example, assuming the primary welding direction is horizontal, a non-uniform angular frequency weight matrix is ​​constructed, with rows corresponding to Zernike moment order n and columns corresponding to angular frequency m. Angular frequency components with a strong correlation to the horizontal direction, i.e., columns with smaller m values, are assigned larger weight coefficients, while other angular frequency components are assigned smaller weight coefficients. The modulus of the composite density-weighted Zernike moments of each order is multiplied by the corresponding weight coefficient in the non-uniform angular frequency weight matrix to obtain the directionally enhanced composite Zernike eigenvector.

[0026] For example, a filter bank consisting of 40 Gabor filters with five center frequencies and eight orientation angles is designed. Each filter in the filter bank is convolved with the weld area sub-image, and the amplitude of the convolution result is taken to generate 40 Gabor energy feature maps. For each of these 40 Gabor energy feature maps, Hu invariant moments from the first to the seventh order are calculated. All calculated Hu invariant moments are sequentially concatenated into a one-dimensional vector with a total dimension of 280, resulting in a multi-scale Gabor-Hu texture feature.

[0027] S4, splicing and fusing the directionally enhanced composite Zernike features with the multi-scale Gabor-Hu texture features to form a high-dimensional composite feature vector, inputting the high-dimensional composite feature vector into a pre-trained random forest classification model, and the random forest classification model outputs the quality detection category of the welding area.

[0028] Specifically, the generated directional enhanced composite Zernike eigenvector is concatenated end-to-end with the generated multi-scale Gabor-Hu texture eigenvector to form a higher-dimensional composite feature vector. This higher-dimensional composite feature vector is fed as input into a random forest classification model trained on a large amount of weld sample data containing multiple category labels, such as acceptable, pore, and crack. The random forest classification model consists of hundreds of decision trees. By combining the voting results of all decision trees, it outputs the quality inspection category of the weld area, such as acceptable or specific defect type.

[0029] In an optional embodiment, the preprocessing of the image to obtain a normalized welding area sub-image includes: converting the acquired image into a grayscale image; using a The grayscale image is filtered using a windowed median filter; the Canny operator is used to perform edge detection on the filtered grayscale image to identify the weld contour; a 256×256 pixel weld area sub-image is cropped from the grayscale image based on the weld contour, and the pixel values ​​of the weld area sub-image are linearly normalized to the interval [0, 1].

[0030] The median filter obtains each pixel and the 24 neighboring pixels around it, a total of 25 pixels, and uses the median of these pixel values ​​to replace the pixel value of the central pixel, which can effectively remove noise while better preserving the clarity of the weld edge. After image denoising, the Canny edge detection algorithm is used to accurately locate the boundary of the weld. The Canny edge detection algorithm can identify areas with drastic brightness changes in the image and generate a clear weld outline. Based on this weld outline, the core position of the weld is determined and a weld is cropped out with this as the center. The weld area sub-image is a pixel-by-pixel image. The grayscale values ​​of all pixels in the weld area sub-image are linearly normalized to the range [0, 1]. Normalization can eliminate the effects of uneven lighting and enhance the adaptability of images taken under different conditions.

[0031] In an optional embodiment, the calculation of the local fractal dimension map of pixels in the gradient magnitude map using the box counting method includes: for each pixel in the gradient magnitude map, taking a pixel as the center An analysis window of pixels; within the analysis window, set a set of boxes of size S {2, 4, 8, 16}; for each box size S, calculate the minimum number of boxes required to cover all non-zero gradient pixels within the analysis window ; In is the horizontal axis, In the double logarithmic coordinate system with y as the ordinate, the least squares linear fitting is performed on the data points, and the absolute value of the slope of the obtained straight line is used as the local fractal dimension of the central pixel.

[0032] For example, for any pixel point on the gradient magnitude map, an analysis window around the pixel point is obtained, such as a The smallest area is used first. The pixel box covers this window, and the number of such boxes needed to cover all pixels with non-zero gradient values ​​is counted, and this number is recorded as N2. Then use a larger box, such as , as well as Repeat the above covering and counting process to get a series of box numbers corresponding to different box sizes, such as N4, N8, N 16 These data points, with the logarithmic value of the reciprocal of the box size as the horizontal coordinate and the logarithmic value of the number of boxes as the vertical coordinate, are approximated as a straight line on the double logarithmic coordinate graph. This line is fitted using the least squares method, and the absolute value of the slope is calculated to obtain the local fractal dimension of the center pixel, which reflects the texture fineness of the local area of ​​the center pixel. This process is repeated for each pixel to generate a complete local fractal dimension map.

[0033] In an optional embodiment, the use of the non-uniform angular frequency weight matrix to adjust the composite density weighted Zernike moments of each order includes: performing Radon transformation on the gradient amplitude map to determine the main welding direction ; Weighted Zernike moments of composite density of each order Multiply by the twiddle factor To achieve rotation normalization; a non-uniform angular frequency weight matrix is ​​used to weight the Zernike moments after rotation normalization.

[0034] In order to make the extracted shape features not affected by the shooting angle of the weld in the image, rotation normalization processing is performed. By performing Radon transform on the gradient amplitude map, the direction of the most significant linear feature in the image is detected, which is determined as the main welding direction. For example, its angle is calculated to be 30 degrees. The composite density weighted Zernike moments of each order are used to calculate the weighted Zernike moments. Multiply by a complex twiddle factor , complex twiddle factors Determined by the angle of the main welding direction.

[0035] After rotational normalization, all welds are aligned to the same datum. However, different Zernike moment components have different importance for describing weld defects. For example, certain high-order moments are more sensitive to describing high-frequency defects such as undercuts. Furthermore, a non-uniform angular frequency weight matrix is ​​used to weight the various Zernike moment components after rotational normalization. For example, moments representing key defect morphology are assigned a weight greater than 1, while moments unrelated to the defect are assigned a weight less than 1. This highlights the most valuable feature information for classification and improves detection accuracy.

[0036] In an optional embodiment, the Gabor filter group consists of 5 center frequencies and 8 direction angles, wherein the center frequencies are {1 / 4, 1 / 8, 1 / 16, 1 / 32, 1 / 64}, the unit of the center frequency is cycle / pixel, and the direction angles are {0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, 157.5°}.

[0037] A filter bank, consisting of 40 independent Gabor filters, is used to capture weld texture information from multiple dimensions. Five center frequencies determine the filter's sensitivity to texture scale. For example, a high-frequency filter like 1 / 4 cycle / pixel can detect very fine and dense texture variations in the image, while a low-frequency filter like 1 / 64 cycle / pixel is more sensitive to large-scale, gentle texture patterns.

[0038] The eight directional angles enable the filter bank to respond to texture features in different directions. These directional angles start at 0 degrees (horizontally) and step 22.5 degrees, covering a range of 180 degrees. This ensures that weld texture details in all directions, whether horizontal, vertical, or diagonal, can be effectively captured. Convolving an image with each of these 40 filters yields 40 energy feature maps, each highlighting the strength of the original image's texture response at a specific scale and orientation, forming a comprehensive weld texture profile.

[0039] In an optional embodiment, the classification model outputs the quality inspection category of the welding area, including: a random forest classification model having 500 decision trees, and the maximum depth of each decision tree is 25, and the node splitting criterion is Gini impurity; the quality inspection category output by the random forest classification model is "qualified", "surface porosity", "incomplete penetration" or "bite".

[0040] The random forest classification model consists of 500 decision trees, analyzing the input feature vector from 500 different perspectives. This results in more accurate analysis than a single decision tree. Each decision tree has a maximum depth of 25 layers, which prevents a single tree from becoming too complex and causing overfitting, ensuring the random forest classification model's ability to generalize to new samples.

[0041] When data needs to be partitioned at a node during the training of each decision tree, Gini impurity is used as an evaluation criterion to select the features and thresholds that yield the best classification results. When the multi-scale Gabor-Hu texture feature vector is input into this trained model, each of the 500 decision trees makes an independent classification decision. The model then uses majority voting to tally the predictions of all decision trees, and the category with the most votes is output as the final quality inspection category.

[0042] The implementation principle of the intelligent detection method for high-precision welding of aircraft aluminum condensers according to the embodiment of the present invention is as follows: by constructing a composite density-weighted Zernike moment, the gradient amplitude reflecting the weld edge information is combined with the local fractal dimension characterizing the complexity of its surface texture, so that the improved Zernike moment can simultaneously describe the macroscopic outline and microscopic details of the weld. Moreover, by introducing a non-uniform angular frequency weight matrix that matches the main direction of welding, the Zernike moment is directionally enhanced and adjusted, thereby improving the sensitivity of the feature to directional defects such as cracks and lack of fusion. In addition, a multi-scale Gabor-Hu texture feature is constructed, and by calculating the Hu invariant moment of the Gabor energy map, detailed features such as the surface roughness of the weld and the weld wave texture are obtained. The directionally enhanced shape feature is fused with the multi-scale invariant texture feature to form a composite feature with strong information complementarity and discrimination, thereby improving the accuracy of detecting complex welding defects in aircraft aluminum condensers.

[0043] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. An intelligent detection method for high-precision welding of aircraft aluminum condensers, characterized in that: The method comprises the following steps: acquiring an image of the welding area of ​​the condenser to be inspected, and preprocessing the image to obtain a normalized sub-image of the welding area; The gradient amplitude map of the welding area sub-image is calculated, and the local fractal dimension map of the pixels in the gradient amplitude map is calculated using the box counting method; the geometric center of the welding area sub-image is taken as the origin, the product of the pixel gradient value in the gradient amplitude map and the corresponding pixel value in the local fractal dimension map is used as the composite density function, and the composite density weighted Zernike moments of each order are calculated; a non-uniform angular frequency weight matrix matching the main welding direction is set, and the non-uniform angular frequency weight matrix is ​​used to adjust the composite density weighted Zernike moments of each order, thereby enhancing the angular frequency component associated with the main welding direction and generating a direction-enhanced composite Zernike moment. Zernike features; a group of Gabor filter groups with different center frequencies and direction angles are used to convolve the welding area sub-image to generate a group of Gabor energy feature maps, seven Hu invariant moments are calculated for each Gabor energy feature map, and all Hu invariant moments are cascaded to form a multi-scale Gabor-Hu texture feature; the direction-enhanced composite Zernike features are spliced ​​and fused with the multi-scale Gabor-Hu texture features to form a high-dimensional composite feature vector, and the high-dimensional composite feature vector is input into a pre-trained classification model, and the classification model outputs the quality detection category of the welding area.

2. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1 is characterized in that: The image is preprocessed to obtain a normalized welding area sub-image, including: converting the acquired image into a grayscale image; using a The grayscale image is filtered using a windowed median filter; the Canny operator is used to perform edge detection on the filtered grayscale image to identify the weld contour; a 256×256 pixel weld area sub-image is cropped from the grayscale image based on the weld contour, and the pixel values ​​of the weld area sub-image are linearly normalized to the interval [0, 1].

3. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1 is characterized in that: The method of calculating the local fractal dimension map of pixels in the gradient amplitude map by using the box counting method includes: for each pixel in the gradient amplitude map, taking a pixel as the center An analysis window of pixels; within the analysis window, set a set of boxes of size S {2, 4, 8, 16}; for each box size S, calculate the minimum number of boxes required to cover all non-zero gradient pixels within the analysis window ; In is the horizontal axis, In the double logarithmic coordinate system with y as the ordinate, the least squares linear fitting is performed on the data points, and the absolute value of the slope of the obtained straight line is used as the local fractal dimension of the central pixel.

4. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1 is characterized in that: The method of adjusting the composite density weighted Zernike moments of each order by using the non-uniform angular frequency weight matrix includes: Perform Radon transform on the gradient amplitude map to determine the main welding direction ; The weighted Zernike moments of the composite density of each order Multiply by the twiddle factor To achieve rotation normalization; The rotation-normalized Zernike moments are weighted using a non-uniform angular frequency weight matrix.

5. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1 is characterized in that: The Gabor filter group consists of 5 center frequencies and 8 direction angles, wherein the center frequencies are {1 / 4, 1 / 8, 1 / 16, 1 / 32, 1 / 64}, the unit of the center frequency is cycle / pixel, and the direction angles are {0°, 22.5°, 45°, 67.5°, 90°, 112.5°, 135°, 157.5°}.

6. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1 is characterized in that: The multi-scale Gabor-Hu texture features are formed, including: For each of the Gabor energy feature maps, calculating Hu invariant moments of order 1 to 7 respectively; All calculated Hu invariant moments are concatenated into a one-dimensional feature vector in a predetermined order.

7. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1 is characterized in that: The classification model is a random forest classification model.

8. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 7, characterized in that: The classification model outputs the quality inspection category of the welding area, including: The random forest classification model has 500 decision trees, and the maximum depth of each decision tree is 25. The node splitting criterion is Gini impurity; The quality inspection categories output by the random forest classification model are "acceptable", "surface porosity", "incomplete penetration", or "undercut".

9. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 1, characterized in that: The method of acquiring an image of the welding area of ​​the condenser to be inspected includes: using a camera in conjunction with a coaxial light source to collect an RGB image of the welding area of ​​the condenser to be inspected.

10. The intelligent detection method for high-precision welding of aircraft aluminum condensers according to claim 9, characterized in that: The preprocessing of the image to obtain a normalized welding area sub-image includes: The RGB image is converted into a grayscale image and smoothed and denoised using a Gaussian filter. The welding area in the grayscale image is cropped and uniformly scaled to 256×256 pixels using a bilinear interpolation algorithm to obtain a welding area sub-image. The pixel values ​​of the welding area sub-image are linearly normalized to the interval [0, 1].

Images