Micro-channel aluminum flat tube appearance detection method based on image processing

Abstract

The invention discloses a micro-channel aluminum flat tube appearance detection method based on image processing, and relates to the technical field of appearance detection.According to the method, more comprehensive information acquisition is carried out through the multispectral imaging technology in combination with RGB, polarized light, infrared spectrum images and ultraviolet spectrum images, and surface temperature difference changes are detected through infrared spectrums; image preprocessing is carried out through contrast-limited adaptive histogram equalization and bilateral filtering, uneven illumination is inhibited while edge details are kept, the contrast of corrosion defects is improved, details of a low-resolution area are enhanced through a super-resolution reconstruction technology, edge features of a corrosion area are further optimized in combination with a histogram in the gradient direction, and the edge features of the corrosion area are further optimized. The identification capability of small-scale defects is improved; an improved YOLOv3 defect detection algorithm is introduced, multi-scale feature extraction is adopted, and a self-attention mechanism is introduced, so that the model pays more attention to a tiny corrosion area, and meanwhile, the detection precision of a small target is enhanced in combination with focus loss and regression loss.

Description

Technical Field

[0001] The present invention relates to the technical field of appearance detection, and in particular to an appearance detection method for microchannel aluminum flat tubes based on image processing. Background Art

[0002] During the large-scale production process of microchannel aluminum flat tubes, their surface quality directly affects the heat exchange performance and service life of the products. Therefore, their appearance detection is extremely important. Currently, on industrial production lines, an automatic detection method based on machine vision is generally adopted, combined with deep learning object detection technology, to automatically identify common defects such as surface scratches and depressions.

[0003] In practical applications, the detection environment of aluminum flat tubes is complex, the material surface has a high reflectivity, and changes in light are likely to cause overexposure or shadows in local areas, making it difficult to identify some defects. Even if polarized light or multi-angle imaging is introduced, it is still difficult to completely eliminate the interference of specular reflection, resulting in the masking of defect information in some highlighted areas.

[0004] In addition, some defects are extremely subtle in the early stage. For example, microcracks caused by stress accumulation during manufacturing or service, or local corrosion points caused by environmental factors. The contrast of these defects in RGB images is extremely low, and traditional YOLO-based detection methods are difficult to stably distinguish between normal areas and defect areas. Especially for corrosion-like defects, their early forms generally show extremely small color differences on the material surface or slight changes in local energy distribution, without forming obvious geometric features. Therefore, there is an urgent need for an appearance detection method for microchannel aluminum flat tubes based on image processing to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] The present invention provides an appearance detection method for microchannel aluminum flat tubes based on image processing to solve the problems that existing RGB vision detection is difficult to distinguish early corrosion, is interfered by reflection, misses tiny cracks, and YOLO is difficult to detect small target defects.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: An embodiment of the present invention provides an appearance detection method for microchannel aluminum flat tubes based on image processing, which includes: Step S1, collecting a collection image of the microchannel aluminum flat tube, using a multi-spectral imaging system, and combining polarized light imaging on the basis of RGB visible light to additionally obtain an infrared spectral image and a ultraviolet spectral image; Step S2, preprocessing the collection image, including light balance, local contrast enhancement, and background suppression, wherein the contrast-limited adaptive histogram equalization (CLAHE) is used to enhance the visibility of corrosion defects; Step S3: Perform super-resolution reconstruction on the acquired image preprocessed in Step S2, and use the super-resolution ESRGAN algorithm to enhance the details in the low-resolution region to generate a super-resolution image; Step S4: Based on the super-resolution image generated in Step S3, adopt the improved YOLOv3 defect detection algorithm to extract the defect region and perform target detection to extract the defect region; Step S5: Classify the defect regions extracted in Step S4, and combine the infrared spectral image, ultraviolet spectral image and visible light information to analyze the temperature anomaly and spectral absorption characteristics of the defect regions through multi-modal data fusion; Step S6: Based on the defect regions classified in Step S5, adopt the dynamic decision threshold method, and optimize the defect decision criteria in combination with historical data to determine the final detection result.

[0008] As a preferred solution of the method for detecting the appearance of micro-channel aluminum flat tubes based on image processing according to the present invention, wherein: the infrared spectral image is used to detect the surface temperature difference.

[0009] As a preferred solution of the method for detecting the appearance of micro-channel aluminum flat tubes based on image processing according to the present invention, wherein: the steps for preprocessing the acquired image are as follows: Perform illumination equalization processing, and perform normalization transformation on the acquired image. The transformation formula is: , wherein, is the gray value of the original acquired image, is the normalized image, is the minimum value of the image pixel value, is the maximum value of the image pixel value, are the horizontal and vertical coordinates of the image, Adopt bilateral filtering to smooth the uneven illumination area. The formula is: , wherein, is the image after bilateral filtering, is the spatial domain Gaussian kernel, is the pixel intensity Gaussian kernel, is the gray value of the neighboring pixel points, are the pixel coordinates within the filtering window, and adopt contrast-limited adaptive histogram equalization for enhancement. The enhancement formula is: , wherein, is the image after CLAHE enhancement, is the contrast-limited adaptive histogram equalization function, is the contrast limit parameter, Calculate the adaptive threshold, and the formula is: , Perform binarization: If , then , If , then , Among them, is the adaptive threshold are the mean and standard deviation within the window respectively, is the adjustment coefficient, is the image after background suppression.

[0010] As a preferred solution of the method for detecting the appearance of a microchannel aluminum flat tube based on image processing according to the present invention, wherein: in step S3, feature enhancement is combined with the histogram of oriented gradients to highlight the edge information of corrosion defects.

[0011] As a preferred solution of the method for detecting the appearance of a microchannel aluminum flat tube based on image processing according to the present invention, wherein: the specific steps of the said step S3 are: Define the downsampling process as: , Among them, is the low-resolution image, is the bicubic interpolation downsampling function, is the image processed in step S2, Perform super-resolution reconstruction using ESRGAN: , Among them, is the image after super-resolution reconstruction, is the ESRGAN generator, is the set of network parameters, Extract the edge information by combining the histogram of oriented gradients HOG, and the extraction formula is: , Among them, is the HOG feature map, is the gradient information along the direction , are the horizontal and vertical coordinates of the image.

[0012] As a preferred solution of the method for detecting the appearance of microchannel aluminum flat tubes based on image processing according to the present invention, in which: during the defect detection process, multi-scale feature extraction is adopted, and a self-attention mechanism is added on the basis of the YOLOv3 algorithm; The focal loss Focal Loss and the regression loss Soft IOU Loss are used as loss functions to enhance the focusing ability on the corrosion defect area during the detection process.

[0013] As a preferred solution of the method for detecting the appearance of microchannel aluminum flat tubes based on image processing according to the present invention, in which: the specific steps of step S4 are as follows: Perform multi-scale feature extraction, and define the multi-scale feature map as : , where respectively represent feature maps of different scales, Calculate the attention weight by using the channel attention mechanism, and the calculation formula is: , where is the attention weight map, is the learnable parameter, is the rectified linear unit activation function, is the Sigmoid activation function, is the pixel point at the feature value, Define the detection box prediction formula as: , where is the target center coordinate, is the target box width and height, is the target confidence, Use Focal Loss for classification optimization, and the optimization function is: , where is the Focal Loss, is the adjustment parameter, is the prediction probability.

[0014] As a preferred solution of the method for detecting the appearance of microchannel aluminum flat tubes based on image processing according to the present invention, in which: the steps of classifying the defect areas extracted in step S4 are as follows: Define the input defect area image as : , Among them, is the defective area image, is the defect mask extracted in step S4, is the super-resolution image of step S3, are the horizontal and vertical coordinates of the image, Calculate the multi-modal feature fusion vector, and the calculation formula is: , Among them, is the multi-modal feature vector, is the visible light feature, is the infrared spectrum feature, is the ultraviolet spectrum feature, is the feature splicing operation, Use a deep neural network DNN for classification, and the classification formula is: , Among them, is the defect category probability vector, is the weight matrix of the classification layer, is the bias vector, is the Softmax function, Calculate the infrared temperature distribution deviation, and the calculation formula is: , Among them, is the temperature anomaly deviation, is the temperature value in the infrared image, is the background average temperature, is the number of pixel points in the defective area.

[0015] As a preferred solution of the method for detecting the appearance of a micro-channel aluminum flat tube based on image processing according to the present invention, wherein: in step S6, for the tiny but potentially expanding corrosion defects, they are marked as key monitoring objects.

[0016] As a preferred solution of the method for detecting the appearance of a micro-channel aluminum flat tube based on image processing according to the present invention, wherein: the specific steps of the said step S6 are: Define the determination threshold as : , Among them, is the dynamic determination threshold, is the mean value of historical defect samples, is the standard deviation of historical defect samples, Q is the adjustment coefficient, Grade the defect severity according to the determination threshold, and the grading method is: If , then , If , then , If , then , Among them, is the defect level, 1 is minor, 2 is medium, and 3 is severe, is the defect severity division coefficient, Define the corrosion defect expansion trend prediction model as: , Among them, is the corrosion defect expansion probability, is the prediction model weight, is the prediction model bias, is the Sigmoid function. If , then it is marked as the key monitoring object, where is the expansion trend determination threshold.

[0017] The beneficial effects of the present invention are as follows: Through multi-spectral imaging technology, combining RGB, polarized light, infrared, and ultraviolet spectral images, more comprehensive information is obtained. The infrared spectrum is used to detect the surface temperature difference change to improve the visibility of the early corrosion area; The restricted contrast adaptive histogram equalization (CLAHE) and bilateral filtering are used for image preprocessing to suppress uneven illumination while maintaining edge details, improving the contrast of corrosion defects and making them more prominent in complex backgrounds.

[0018] The present invention uses super-resolution reconstruction technology to enhance the details of low-resolution areas, combines the histogram of oriented gradients to further optimize the edge features of the corrosion area, and improves the recognition ability of small-scale defects; An improved YOLOv3 defect detection algorithm is introduced, which uses multi-scale feature extraction and introduces a self-attention mechanism to make the model pay more attention to small corrosion areas. At the same time, focal loss and regression loss are combined to enhance the detection accuracy of small targets and avoid the missed detection problem caused by large targets dominating the training.

[0019] For defect classification, the present invention adopts multi-modal data fusion, stitches infrared, ultraviolet, and visible light features into a complete feature vector, and classifies them through a deep neural network. At the same time, infrared temperature deviation analysis is combined to further improve the classification accuracy of corrosion defects; Combining historical data, the dynamic determination threshold method is used to optimize the defect determination standard, and the expansion trend prediction model is used to highlight small but potentially developing corrosion defects for subsequent quality control and maintenance.

[0020] In summary, the present invention effectively improves the accuracy of the appearance detection of microchannel aluminum flat tubes, can timely detect early corrosion defects, reduce missed detections and false detections, and improve the detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0022] Figure 1 It is a schematic flowchart of the appearance detection method for microchannel aluminum flat tubes based on image processing of the present invention.

[0023] Figure 2 It is an image of the microchannel aluminum flat tube after preprocessing in step S2.

[0024] Figure 3 It is an image of another microchannel aluminum flat tube after preprocessing in step S2.

[0025] Figure 4 It is Figure 2 an image processed in step S3. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] In order to make the above objects, features and advantages of the present invention more obvious and understandable, the following will give a detailed description of the specific embodiments of the present invention with reference to the drawings in the specification.

[0027] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0028] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" appearing in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0029] Embodiment 1, referring to Figure 1 , this embodiment provides an appearance detection method for microchannel aluminum flat tubes based on image processing, including the following steps: Step S1, collect the acquisition image of the microchannel aluminum flat tube, use a multi-spectral imaging system, and combine polarized light imaging on the basis of RGB visible light to additionally obtain infrared spectral images and ultraviolet spectral images; Infrared spectral images are used to detect surface temperature differences.

[0030] Step S2: Preprocess the acquired images, including illumination equalization, local contrast enhancement, and background suppression, where Contrast Limited Adaptive Histogram Equalization (CLAHE) is used for equalization to enhance the visibility of corrosion defects. The steps for preprocessing the acquired images are as follows: Perform illumination equalization processing. The acquired images are subjected to normalization transformation, and the transformation formula is: , where, is the gray value of the original acquired image, is the normalized image, is the minimum value of the image pixel values, is the maximum value of the image pixel values, are the horizontal and vertical coordinates of the image, Use bilateral filtering to smooth the uneven illumination area. The formula is: , where, is the image after bilateral filtering, is the Gaussian kernel in the spatial domain, is the Gaussian kernel of pixel intensity, is the gray value of the neighboring pixel points, are the pixel coordinates within the filtering window, Use Contrast Limited Adaptive Histogram Equalization for enhancement. The enhancement formula is: , where, is the image after CLAHE enhancement, is the Contrast Limited Adaptive Histogram Equalization function, is the contrast limit parameter, Calculate the adaptive threshold. The formula is: , Perform binarization: If , then , If , then , where, is the adaptive threshold, are the mean and standard deviation within the window respectively, is the adjustment coefficient, is the image after background suppression, Specifically, here the influence of illumination variation is reduced through a normalization method to stabilize the brightness distribution of the image; bilateral filtering is adopted to balance the uneven illumination in the image, improve the overall uniformity, and at the same time retain the edge detail information; contrast-limited adaptive histogram equalization is used to enhance the local contrast, making the tiny corrosion defects more obvious; and background suppression is carried out through an adaptive threshold method to reduce the interference of irrelevant information, thereby retaining the important defect areas.

[0031] Step S3: Based on the acquired image preprocessed in step S2, super-resolution reconstruction is performed. The details of the low-resolution region are enhanced using the super-resolution ESRGAN algorithm to generate a super-resolution image.

[0032] In step S3, feature enhancement is combined with the histogram of oriented gradients to highlight the edge information of the corrosion defects; The steps for performing super-resolution reconstruction on the acquired image preprocessed in step S2, using the super-resolution ESRGAN algorithm to enhance the details of the low-resolution region and generate a super-resolution image are as follows: Define the downsampling process as: , where, is the low-resolution image, is the bicubic interpolation downsampling function, is the image processed in step S2, Perform super-resolution reconstruction using ESRGAN: , where, is the image after super-resolution reconstruction, is the ESRGAN generator, is the set of network parameters, Combine the histogram of oriented gradients HOG to extract the edge information. The extraction formula is: , where, is the HOG feature map, is the gradient information along the direction , are the horizontal and vertical coordinates of the image.

[0033] Specifically, here the super-resolution generative adversarial network is used to reconstruct the low-resolution image, enhance the image details, and make the defect areas clearer; during the reconstruction process, perceptual loss, adversarial loss, and content loss are jointly optimized, so that the generated super-resolution image not only has high visual quality but also retains sufficient texture details.

[0034] Step S4: Based on the super-resolution image generated in step S3, use the improved YOLOv3 defect detection algorithm to extract the defect area and perform target detection to extract the defect area; During the defect detection process, multi-scale feature extraction is adopted, and a self-attention mechanism is added based on the YOLOv3 algorithm; Use the Focal Loss and the regression loss Soft IOU Loss as loss functions to enhance the focusing ability on the corrosion defect area during the detection process.

[0035] The steps of using the improved YOLOv3 defect detection algorithm to extract the defect area and perform target detection based on the super-resolution image generated in step S3 are as follows: Perform multi-scale feature extraction, and define the multi-scale feature map as : , where represent feature maps of different scales respectively, Calculate the attention weight using the channel attention mechanism, and the calculation formula is: , where is the attention weight map, is the learnable parameter, is the rectified linear unit activation function, is the Sigmoid activation function, is the feature value at pixel point , Define the detection box prediction formula as: , where is the target center coordinate, is the width and height of the target box, is the target confidence, Use Focal Loss for classification optimization, and the optimization function is: , where is the Focal Loss, is the adjustment parameter, is the prediction probability, Specifically, this step adopts a multi-scale feature extraction method to capture defect features at different scales, improving the recognition ability of tiny defects; introduces a channel attention mechanism to enhance the network's focusing ability on the defect area; uses Focal Loss to optimize the classification loss, enhancing the model's learning ability for difficult-to-detect defects, and at the same time uses a regression loss to improve the accuracy of the target box. Combining super-resolution images and deep learning detection algorithms to improve the recognition efficiency of corrosion defects.

[0036] Step S5: Classify the defect areas extracted in step S4, and combine infrared spectral images, ultraviolet spectral images and visible light information to analyze the temperature anomaly and spectral absorption characteristics of the defect areas through multi-modal data fusion. The steps for classifying the defect areas extracted in step S4 are as follows: Define the input defect area image as : , where is the defect area image, is the defect mask extracted in step S4, is the super-resolution image of step S3, are the horizontal and vertical coordinates of the image, Calculate the multi-modal feature fusion vector, and the calculation formula is: , where is the multi-modal feature vector, is the visible light feature, is the infrared spectral feature, is the ultraviolet spectral feature, is the feature splicing operation, Use a deep neural network DNN for classification, and the classification formula is: , where is the defect category probability vector, is the classification layer weight matrix, is the bias vector, is the Softmax function, Calculate the infrared temperature distribution deviation, and the calculation formula is: , where is the temperature anomaly deviation, is the temperature value in the infrared image, is the background average temperature, is the number of pixels in the defect area, Specifically, based on the defect regions extracted in step S4, the information of visible light, infrared spectrum, and ultraviolet spectrum is merged through a multimodal data fusion method to form a complete feature vector. A deep neural network is used to classify the fused data to determine the probability distribution of defect categories. The severity of the defect is determined by combining temperature anomaly analysis, making the classification process more accurate.

[0037] Step S6: Based on the defect regions classified in step S5, the dynamic decision threshold method is adopted, and the defect determination criteria are optimized by combining historical data to determine the final detection result. In step S6, for the tiny but potentially expanding corrosion defects, they are marked as key monitoring objects. The so-called tiny defects refer to the defects with the number of pixel points less than a certain threshold, and this threshold is specifically defined according to the size of the microchannel aluminum flat tube and the use of the product.

[0038] The steps of adopting the dynamic decision threshold method, combining historical data to optimize the defect determination criteria, and determining the final detection result based on the defect regions classified in step S5 are as follows: Define the decision threshold as : , where is the dynamic decision threshold, is the mean of the historical defect samples, is the standard deviation of the historical defect samples, is the adjustment coefficient, Classify the defect severity according to the decision threshold, and the classification method is: If , then , If , then , If , then , where is the defect level, 1 is minor, 2 is medium, 3 is severe, is the defect severity division coefficient, Define the corrosion defect expansion trend prediction model as: , where is the corrosion defect expansion probability, is the prediction model weight, is the prediction model bias, is the Sigmoid function. If , then it is marked as a key monitoring object, where is the expansion trend decision threshold; Specifically, dynamic thresholds are calculated based on historical data to make the defect determination process more intelligent and able to adapt to the detection needs of different working conditions. The defect level is calculated based on the temperature abnormality deviation, and for corrosion defects that may expand, a prediction model is introduced to calculate the probability of corrosion expansion to determine its development trend, which is convenient for subsequent maintenance.

[0039] Figure 2 and Figure 3 All of them are images of microchannel aluminum flat tubes after preprocessing in step S2. From the images, it can be seen that the microchannel aluminum flat tubes have quality defects, but there are still cases of missed inspections. Figure 4 ,The image details after processing in step S3 are clearer, which is convenient for subsequent extraction of defective areas, such as Figure 4 The boxed area in the original image is not obviously visible, but it can be accurately extracted after processing, which can greatly improve the detection accuracy and avoid missed detection.

[0040] Super-resolution reconstruction technology is used to enhance the details of low-resolution areas, and the edge features of the corrosion area are further optimized in combination with the gradient direction histogram to improve the recognition ability of small-scale defects. The improved YOLOv3 defect detection algorithm is introduced, multi-scale feature extraction is adopted, and the self-attention mechanism is introduced to make the model pay more attention to the tiny corrosion area. At the same time, the focus loss and regression loss are combined to enhance the detection accuracy of small targets and avoid the problem of missed detection caused by large target-dominated training. Combined with historical data, the dynamic judgment threshold method is used to optimize the defect judgment standard, and the small but possible corrosion defects are marked by expanding the trend prediction model for subsequent quality control and maintenance.

[0041] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for detecting the appearance of a microchannel aluminum flat tube based on image processing, characterized in that: include, Step S1, collecting images of the microchannel aluminum flat tube, using a multi-spectral imaging system, combining polarized light imaging on the basis of RGB visible light, and additionally obtaining infrared spectrum images and ultraviolet spectrum images; Step S2, preprocessing the collected image, including illumination equalization, local contrast enhancement and background suppression, wherein limited contrast adaptive histogram (CLAHE) equalization is used; Step S3, performing super-resolution reconstruction based on the collected image pre-processed in step S2, using the super-resolution ESRGAN algorithm to enhance the details of the low-resolution area to generate a super-resolution image; Step S4, based on the super-resolution image generated in step S3, using an improved YOLOv3 defect detection algorithm, extracting defect areas and performing target detection to extract defect areas; Step S5, classifying the defective areas extracted in step S4, combining the infrared spectrum image, the ultraviolet spectrum image and the visible light information, and analyzing the temperature anomaly and spectral absorption characteristics of the defective areas through multimodal data fusion; Step S6, based on the defect area classified in step S5, a dynamic determination threshold method is used to optimize the defect determination criteria in combination with historical data to determine the final detection result.

2. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 1, characterized in that: The infrared spectrum image is used to detect surface temperature differences.

3. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 1 is characterized in that: The step of preprocessing the collected image is: Perform illumination balance processing and collect images for normalization transformation. The conversion formula is: ， in, is the gray value of the original acquired image, is the normalized image, is the minimum value of the image pixel value, is the maximum value of the image pixel value, are the horizontal and vertical coordinates of the image; Bilateral filtering is used to smooth uneven illumination areas. The formula is: ， in, is the bilaterally filtered image, is the Gaussian kernel in the spatial domain, is the pixel intensity Gaussian kernel, is the gray value of the neighborhood pixel, is the pixel coordinate within the filter window; The contrast-limited adaptive histogram equalization is used for enhancement, and the enhancement formula is: ， in, is the image enhanced by CLAHE. To limit the contrast of the adaptive histogram equalization function, is the contrast limit parameter; Calculate the adaptive threshold, the formula is: ， Binarization: like ,but , like ,but , in, is the adaptive threshold, Window The mean and standard deviation within is the adjustment coefficient, is the image after background suppression.

4. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 1 is characterized in that: In step S3, feature enhancement is performed in combination with the gradient direction histogram to highlight the edge information of the corrosion defect.

5. The microchannel aluminum flat tube appearance detection method based on image processing as claimed in claim 4 is characterized by: The specific steps of step S3 are: The downsampling process is defined as: ， in, For low-resolution images, is the bicubic interpolation downsampling function, is the image processed in step S2, Super-resolution reconstruction using ESRGAN: ， in, is the image after super-resolution reconstruction, is the ESRGAN generator, is a set of network parameters; Combined with the gradient directional histogram HOG, the edge information is extracted, and the extraction formula is: ， in, is the HOG feature map, Along the direction The gradient information of are the horizontal and vertical coordinates of the image.

6. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 1, characterized in that: In the defect detection process of step S4, multi-scale feature extraction is adopted, and a self-attention mechanism is added on the basis of the YOLOv3 algorithm; Focal Loss and Soft IOU Loss are used as loss functions to enhance the focusing ability on the corrosion defect area during the detection process.

7. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 6 is characterized in that: The specific steps of step S4 are: Perform multi-scale feature extraction and define the multi-scale feature map as : ， in, They represent feature maps of different scales, The channel attention mechanism is used to calculate the attention weight, and the calculation formula is: ， in, is the attention weight map, is a learnable parameter, is the rectified linear unit activation function, is the Sigmoid activation function, Pixel The eigenvalue at ; The detection box prediction formula is defined as: ， in, is the target center coordinate, is the target box width and height, is the target confidence, FocalLoss is used for classification optimization, and the optimization function is: ， in, is Focal Loss, To adjust the parameters, is the predicted probability.

8. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 7 is characterized in that: The steps of classifying the defective areas extracted in step S4 are as follows: Define the input defect area image as : ， in, is the defect area image, is the defect mask extracted in step S4, is the super-resolution image of step S3, are the horizontal and vertical coordinates of the image; Calculate the multimodal feature fusion vector, the calculation formula is: ， in, is the multimodal feature vector, is the visible light characteristic, is the infrared spectral characteristic, is the ultraviolet spectral characteristic, It is a feature concatenation operation; The deep neural network DNN is used for classification, and the classification formula is: ， in, is the defect category probability vector, is the classification layer weight matrix, is the bias vector, is the Softmax function; Calculate the infrared temperature distribution deviation, the calculation formula is: ， in, is the temperature abnormal deviation, is the temperature value in the infrared image, is the background average temperature, is the number of pixels in the defect area.

9. The microchannel aluminum flat tube appearance detection method based on image processing according to claim 1, characterized in that: In step S6, small corrosion defects with potential expansion tendency are marked as key monitoring objects.

10. A method for detecting the appearance of a microchannel aluminum flat tube based on image processing according to claim 1 or 9, characterized in that: The specific steps of step S6 are: The decision threshold is defined as : ， in, is the dynamic decision threshold, is the mean of historical defect samples, is the standard deviation of historical defect samples, is the adjustment coefficient, The severity of defects is graded according to the judgment threshold. The grading method is: like ,but , like ,but , like ,but , in, is the defect level, 1 is slight, 2 is moderate, 3 is severe, is the defect severity classification coefficient; The corrosion defect expansion trend prediction model is defined as: ， in, is the corrosion defect expansion probability, is the prediction model weight, is the prediction model bias, is the Sigmoid function, if , it is marked as a key monitoring object, among which The threshold for determining the extension trend.

Images