A defect detection method for automobile exhaust device

Abstract

The present invention belongs to the field of computer image processing, and specifically relates to a defect detection method for automobile exhaust devices. It includes multi-angle image acquisition, quantum-inspired enhanced image, multi-modal fusion feature extraction, and construction of a dynamic standard feature model for comparison and judgment of defects and positions. The method improves image quality, can extract features more comprehensively and accurately, can effectively detect a variety of defects, and adapts to different situations through a dynamic standard feature model, thereby improving the accuracy and efficiency of detection, overcoming the problems of low efficiency and strong subjectivity of manual detection in the prior art, and poor effect of machine vision detection of complex defects, thereby ensuring the quality and performance of automobile exhaust devices.

Description

Technical Field

[0001] The invention belongs to the field of computer image processing, and in particular relates to a defect detection method for an automobile exhaust device. Background Art

[0002] With the rapid development of the automobile industry, the impact of automobile exhaust emissions on the environment has received increasing attention. As a key component for controlling exhaust emissions, the quality and performance of automobile exhaust devices are directly related to the compliance of exhaust emissions. However, during the production process of automobile exhaust devices, defects such as scratches, dents, and bulges may occur due to various reasons. These defects will not only affect the appearance of the exhaust device, but also its performance and service life. At present, the detection methods for defects in automobile exhaust devices mainly include manual detection and traditional machine vision detection. Manual detection is inefficient, and the detection results are easily affected by the subjective factors of the inspectors. Although the existing machine vision detection methods have improved the detection efficiency and accuracy to a certain extent, the detection effect of complex defects such as scratches and dents is still not ideal. In addition, the existing detection methods can often only detect a single type of defect, and cannot detect multiple types of defects at the same time, such as scratches, dents, bulges, and heat-related defects. Moreover, for the detection of defects in automobile exhaust devices, there is a lack of a method that can dynamically adjust the detection standard according to the actual situation to adapt to different production environments and product requirements. Therefore, there is a need for a new, efficient, accurate automobile exhaust device defect detection method that can simultaneously detect multiple types of defects and dynamically adjust the detection standards. Summary of the invention

[0003] In view of the above technical problems existing in the detection of automobile exhaust devices, the present invention proposes a defect detection method for automobile exhaust devices using machine vision image processing.

[0004] In order to achieve the above object, the technical solution adopted by the present invention comprises the following steps:

[0005] S1. First, use an image acquisition device to acquire images of the automobile exhaust device from multiple angles to obtain multiple images to be detected;

[0006] S2. Secondly, a quantum-inspired image enhancement method is applied to the collected image to be detected to improve the quality of the image data;

[0007] S3, then using a feature extraction method based on multimodal fusion to extract feature information in the image to be detected, the feature information including scratches, concave and convex detection and heat-related features;

[0008] S4. Finally, a dynamic standard feature model is constructed, and an intelligent comparison algorithm is used to compare the extracted feature information with the preset defect-free standard features to determine whether the automobile exhaust device has defects; if there are defects, the defect location is output.

[0009] Preferably, the specific operation of step S1 in the image acquisition stage is:

[0010] S11. First, a high-resolution industrial camera is selected as the image acquisition device. According to the shape and size of the automobile exhaust device, the installation position and angle of the camera are determined to ensure that the automobile exhaust device can be imaged from multiple angles and in all directions. The installation position is based on Where D is the distance between the camera and the center of the vehicle exhaust device, L is the length of the vehicle exhaust device, W is the width, and H is the height;

[0011] S12. Secondly, it is necessary to set appropriate lighting conditions. The lighting intensity is adjusted according to the material and characteristics of the automobile exhaust device to ensure that the defect features in the image can be clearly displayed. The relationship between the lighting intensity I and the reflectivity R of the automobile exhaust device is satisfied. Where α is the adjustment coefficient;

[0012] S13. Finally, it is necessary to control the camera's shooting parameters, including exposure time and focal length, to obtain high-quality images to be detected. The exposure time T is determined according to the ambient light intensity E and the camera sensitivity S, satisfying the formula Where β is an empirical coefficient, and the focal length f is determined according to the size of the automobile exhaust device and the shooting distance, so that the automobile exhaust device can be clearly presented in the image while avoiding image distortion.

[0013] Preferably, the specific implementation process of the quantum-inspired image enhancement method in step S2 is:

[0014] S21. First, the gray value g(x,y) of each pixel of the automobile exhaust device image is converted into the quantum state amplitude A(x,y). The formula is: Where (x, y) represents the coordinates of the pixel, the grayscale value ranges from 0 to 255, and ω and ε are adjustment parameters used to introduce additional changes and fluctuations to increase the diversity of quantum states;

[0015] S22, then calculate the grayscale difference D(x,y) between the pixel and its neighboring pixels, the formula is Where N(x,y) represents the neighborhood of the pixel, dist((x,y),(i,j)) represents the distance between the pixels (x,y) and (i,j), and λ is the adjustment parameter to control the weight of the influence of distance on grayscale difference;

[0016] S23, then adjust the quantum state amplitude according to the grayscale difference, the formula is Among them, α is the contrast enhancement parameter, μ is the adjustment parameter, It is the average value of the quantum state amplitude of all pixels in the image;

[0017] S24, then perform adaptive quantum color adjustment, calculate the distribution probability P(x, y) of the color of the pixel in the entire image color space, and adjust the quantum state amplitude A′(x, y)=A(x, y)×(1+β×P(x, y))×ρ(x, y) by counting the histograms of different color channels in the image and calculating the probability according to the position of the color value of the pixel in the histogram, where β is a color adjustment degree parameter and ρ(x, y) is a weight factor dynamically adjusted according to the spatial position and color characteristics of the pixel;

[0018] S25. In addition, multi-scale quantum texture enhancement is performed, and the texture information T(x, y) of the image is extracted using multi-scale wavelet transform, and the details and structure of the texture are analyzed at different scales, and the quantum state amplitude A′(x, y)=A(x, y)×(1+γ×T(x, y))×σ(x, y) is adjusted, where γ is a texture enhancement degree parameter, and σ(x, y) is a weight factor dynamically adjusted according to the scale and direction of the texture;

[0019] S26. Finally, the quantum state amplitude after quantum evolution is converted back to the traditional image representation, that is, g(x,y)=255×A′(x,y) 2 Get the enhanced image.

[0020] Preferably, the specific implementation of extracting feature information from the image to be detected by using a feature extraction method based on multimodal fusion in step S3 is:

[0021] S31, first extract scratches, depressions and protrusions. The specific implementation is to use the image and depth information fusion method to extract the feature method, which is manifested as changes in color texture in the image. The feature is F img , and the depth information may show depth changes, characterized by F depth , then the comprehensive feature is F1=α1F img +α2F depth , where α1 and α2 are weight coefficients, which evaluate the results under different weight combinations through machine learning and select the optimal weight combination;

[0022] S32, then extract the heat-related features, infrared-assisted research is used for the heat-related features, which are manifested as changes in temperature distribution in the infrared image. The characteristics of the heat features in the infrared image are F heat_ir , the characteristic of thermal feature in visible light image is F heat_vis , then the comprehensive feature is F2=β1F heat_ir +β2Fheat_vis , where β1, β2 are weight coefficients;

[0023] S33, finally, the scratch, concave, and convex features obtained by fusing the image with the depth information and the thermal related features obtained by fusing the image with the infrared information are further fused to obtain the final comprehensive feature F total =ω1F1+ω2F2, where ω1, ω2 are weight coefficients.

[0024] Preferably, in step S4, a dynamic standard feature model is constructed, and the extracted feature information is compared with a preset defect-free standard feature using an intelligent comparison algorithm to determine whether the automobile exhaust device has defects. The implementation steps are:

[0025] S41, firstly collect images and depth information of multiple defect-free automobile exhaust devices, as well as corresponding infrared images, to construct an initial defect-free standard feature set S0;

[0026] S42, then the dynamic update mechanism is implemented to continuously add the features of the newly detected defect-free samples to the standard feature set, and the standard feature set is optimized and updated regularly. updated =S0∪F new -{F outdated}, where F new is the newly detected defect-free sample feature, {F outdated} are features that have been judged to be outdated and inaccurate;

[0027] S43. Finally, the similarity between the extracted features of the automobile exhaust device to be detected and the features in the standard feature set is calculated by the mean absolute error from scratches, concave and convex features and heat-related features, and the similarities of multiple dimensions are integrated to obtain an overall similarity index, the formula is S=γ scratch S scratch +γ dent S dent +γ bump S bump +γ heat S heat , if the overall similarity is lower than the threshold value of 0.75, it is judged that the automobile exhaust device has defects; for the automobile exhaust device with defects, the specific location of the defect is determined by analyzing the position information in the extracted features.

[0028] Compared with the prior art, the advantages and positive effects of the present invention are that quantum-inspired image enhancement is used to improve image quality, multimodal fusion feature extraction is more comprehensive and accurate, the dynamic standard feature model and intelligent comparison algorithm can accurately detect a variety of defects with high efficiency, can adapt to different situations, and ensure the quality of the exhaust device. DETAILED DESCRIPTION

[0029] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0030] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0031] Embodiment, with the vigorous development of the automobile industry, the impact of automobile exhaust emissions on the environment has become increasingly prominent, and the quality and performance of automobile exhaust devices are of vital importance. However, during the production process, various defects may occur in the exhaust device, such as scratches, dents, protrusions, etc. These defects not only affect the appearance, but also may affect its function and service life. Therefore, this embodiment aims to demonstrate the application process and effect of a new, efficient and accurate method for detecting surface defects of automobile exhaust devices. First of all, considering the problem of image collection, for image processing problems, the first thing that needs to be guaranteed is the quality of image capture. Therefore, the present invention uses a high-resolution industrial camera as the image acquisition device. For a specific automobile exhaust device, its length is 50 cm, width is 30 cm, and height is 20 cm, According to the installation position formula, the distance D between the camera and the center of the vehicle exhaust device is set to 80cm, and the camera installation angle ensures that the exhaust device can be imaged from multiple angles and in all directions. α is 0.8, and the appropriate light intensity is determined by the formula; finally, the camera shooting parameters need to be controlled, and the exposure time T is determined according to the ambient light intensity E and the camera sensitivity S, satisfying the formula Where β is an empirical coefficient, the focal length f is determined according to the size of the car exhaust device and the shooting distance, the ambient light intensity is 500, the camera sensitivity S is 400, and β is 0.01 to calculate the exposure time.

[0032] Then, considering the need to enhance the graphics and improve the image quality, the quantum-inspired image enhancement method can convert the pixel grayscale value into quantum state amplitude and better adjust the image through quantum characteristics. Compared with common methods, it can handle pixel differences more finely. First, the grayscale value g(x,y) of each pixel of the automobile exhaust device image is converted into the quantum state amplitude A(x,y), and the formula is Where (x, y) represents the coordinates of the pixel, the grayscale value ranges from 0 to 255, ω and ε are adjustment parameters used to introduce additional changes and fluctuations to increase the diversity of quantum states; secondly, the grayscale difference D(x, y) between the pixel and its neighboring pixels is calculated, and the formula is Where N(x,y) represents the neighborhood of the pixel, dist((x,y),(i,j)) represents the distance between the pixels (x,y) and (i,j), and λ is the adjustment parameter to control the weight of the influence of distance on the grayscale difference. Then, the quantum state amplitude is adjusted according to the grayscale difference. The formula is: Among them, α is the contrast enhancement parameter, μ is the adjustment parameter, is the average value of the quantum state amplitude of all pixels in the image; then adaptive quantum color adjustment is performed to calculate the distribution probability P(x, y) of the pixel color in the entire image color space, and the quantum state amplitude A′(x, y) = A(x, y) × (1 + β × P(x, y)) × ρ(x, y) is adjusted by statistically analyzing the histograms of different color channels in the image and calculating the probability based on the position of the pixel color value in the histogram, where β is the color adjustment degree parameter and ρ(x, y) is a weight factor that is dynamically adjusted based on the spatial position and color characteristics of the pixel. In addition, multi-scale quantum texture enhancement is performed, and the texture information T(x,y) of the image is extracted using multi-scale wavelet transform. The details and structure of the texture are analyzed at different scales, and the quantum state amplitude A′(x,y)=A(x,y)×(1+γ×T(x,y))×σ(x,y) is adjusted, where γ is the texture enhancement parameter, and σ(x,y) is a weight factor that is dynamically adjusted according to the scale and direction of the texture. Finally, the quantum state amplitude after quantum evolution is converted back to the traditional image representation, that is, g(x,y)=255×A′(x,y) 2 Get the enhanced image.

[0033] In order to better extract parameter features in the image, the present invention adopts a feature extraction method based on multimodal fusion to extract feature information in the image to be detected. This method fuses the image with depth, infrared and other information to fully capture features. It can detect a variety of defects, such as scratches, depressions, protrusions and heat-related features. Compared with a single mode, it improves the accuracy and reliability of detection and is more suitable for complex detection needs. First, the scratch, depression and protrusion features are extracted, and the feature extraction method is performed by fusing the image and depth information. The feature is expressed as a change in color texture in the image, and the feature is F img , and the depth information may show depth changes, characterized by F depth , then the comprehensive feature is F1=α1F img +α2F depth, where α1 and α2 are weight coefficients. The results under different weight combinations are evaluated through machine learning to select the optimal weight combination. Then, the thermal related features are extracted. Infrared-assisted research is used for thermal related features. The changes in temperature distribution are shown in infrared images. The characteristics of thermal features in infrared images are F heat_ir , the characteristic of thermal feature in visible light image is F heat_vis , then the comprehensive feature is F2=β1F heat_ir +β2F heat_vis , where β1 and β2 are weight coefficients; finally, the scratch, concave, and convex features obtained by fusion of the image and depth information and the thermal related features obtained by fusion of the image and infrared information are further fused to obtain the final comprehensive feature F total =ω1F1+ω2F2, where ω1, ω2 are weight coefficients.

[0034] After extracting feature information in the above process, a dynamic standard feature model is finally constructed, and dynamic updates are used to adapt to production changes and improve standard accuracy. The intelligent comparison algorithm can also perform multi-dimensional analysis, and the comprehensive judgment is more accurate. It can effectively detect defects, locate more accurately, improve detection efficiency and quality, and ensure the performance of automobile exhaust devices. First, the images and depth information of multiple defect-free automobile exhaust devices, as well as the corresponding infrared images, are collected to construct an initial defect-free standard feature set S0; then a dynamic update mechanism is implemented to continuously add the features of newly detected defect-free samples to the standard feature set, and the standard feature set is regularly optimized and updated. S updated =S0∪F new -{F outdated}, where F new is the newly detected defect-free sample feature, {F outdated} is a feature that is judged to be outdated and inaccurate; finally, the similarity between the features of the extracted automobile exhaust device to be detected and the features in the standard feature set is calculated by the mean absolute error from scratches, dents and convexities, and heat-related features, and the similarities of multiple dimensions are combined to obtain an overall similarity index, the formula is S = γ scratch S scratch +γ dent S dent +γ bump S bump +γ heat S heat , if the overall similarity is lower than the threshold value of 0.75, it is judged that the automobile exhaust device has defects; for the defective automobile exhaust device, the specific location of the defect is determined by analyzing the position information in the extracted features.

[0035] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A defect detection method for an automobile exhaust device, characterized in that: The following steps are involved: S1. First, use an image acquisition device to acquire images of the automobile exhaust device from multiple angles to obtain multiple images to be detected; S2, secondly, the collected image to be detected is subjected to a quantum-inspired image enhancement method to improve the quality of the image data. The specific implementation process of the quantum-inspired image enhancement method in step S2 is as follows: S21. First, the gray value g(x,y) of each pixel of the automobile exhaust device image is converted into the quantum state amplitude A(x,y). The formula is: Where (x, y) represents the coordinates of the pixel, the grayscale value ranges from 0 to 255, and ω and ε are adjustment parameters used to introduce additional changes and fluctuations to increase the diversity of quantum states; S22, then calculate the grayscale difference D(x,y) between the pixel and its neighboring pixels, the formula is Where N(x,y) represents the neighborhood of the pixel, dist((x,y),(i,j)) represents the distance between the pixels (x,y) and (i,j), and λ is the adjustment parameter to control the weight of the influence of distance on grayscale difference; S23, then adjust the quantum state amplitude according to the grayscale difference, the formula is Among them, α is the contrast enhancement parameter, μ is the adjustment parameter, It is the average value of the quantum state amplitude of all pixels in the image; S24, then perform adaptive quantum color adjustment, calculate the distribution probability P(x, y) of the color of the pixel in the entire image color space, and adjust the quantum state amplitude A′(x, y)=A(x, y)×(1+β×P(x, y))×ρ(x, y) by counting the histograms of different color channels in the image and calculating the probability according to the position of the color value of the pixel in the histogram, where β is a color adjustment degree parameter and ρ(x, y) is a weight factor dynamically adjusted according to the spatial position and color characteristics of the pixel; S25. In addition, multi-scale quantum texture enhancement is performed, and the texture information T(x, y) of the image is extracted using multi-scale wavelet transform, and the details and structure of the texture are analyzed at different scales, and the quantum state amplitude A′(x, y)=A(x, y)×(1+γ×T(x, y))×σ(x, y) is adjusted, where γ is a texture enhancement degree parameter, and σ(x, y) is a weight factor dynamically adjusted according to the scale and direction of the texture; S26. Finally, the quantum state amplitude after quantum evolution is converted back to the traditional image representation, that is, g(x,y)=255×A′(x,y) 2 Obtain an enhanced image; S3, then using a feature extraction method based on multimodal fusion to extract feature information in the image to be detected, the feature information including scratches, concave and convex detection and heat-related features; The specific implementation of extracting feature information from the image to be detected by using the feature extraction method based on multimodal fusion in step S3 is: S31, first extract scratches, depressions and protrusions. The specific implementation is to use the image and depth information fusion method to extract the feature method, which is manifested as changes in color texture in the image. The feature is F img , and the depth information may show depth changes, characterized by F depth , then the comprehensive feature is F1=α1F img +α2F depth , where α1 and α2 are weight coefficients, which evaluate the results under different weight combinations through machine learning and select the optimal weight combination; S32, then extract the heat-related features, infrared-assisted research is used for the heat-related features, which are manifested as changes in temperature distribution in the infrared image. The characteristics of the heat features in the infrared image are F heat_ir , the characteristic of thermal feature in visible light image is F heat_vis , then the comprehensive feature is F2=β1F heat_ir +β2F heat_vis , where β1, β2 are weight coefficients; S33, finally, the scratch, concave, and convex features obtained by fusing the image with the depth information and the thermal related features obtained by fusing the image with the infrared information are further fused to obtain the final comprehensive feature F total =ω1F1+ω2F2, where ω1, ω2 are weight coefficients; S4. Finally, a dynamic standard feature model is constructed, and an intelligent comparison algorithm is used to compare the extracted feature information with the preset defect-free standard features to determine whether the automobile exhaust device has defects; if there are defects, the defect location is output.

2. A defect detection method for automobile exhaust device according to claim 1, characterized in that: The specific operations of step S1 in the image acquisition stage are: S11. First, a high-resolution industrial camera is selected as the image acquisition device. According to the shape and size of the automobile exhaust device, the installation position and angle of the camera are determined to ensure that the automobile exhaust device can be imaged from multiple angles and in all directions. The installation position is based on Where D is the distance between the camera and the center of the vehicle exhaust device, L is the length of the vehicle exhaust device, W is the width, and H is the height; S12. Secondly, it is necessary to set appropriate lighting conditions. The lighting intensity is adjusted according to the material and characteristics of the automobile exhaust device to ensure that the defect features in the image can be clearly displayed. The relationship between the lighting intensity I and the reflectivity R of the automobile exhaust device is satisfied. Where α is the adjustment coefficient; S13. Finally, it is necessary to control the camera's shooting parameters, including exposure time and focal length, to obtain high-quality images to be detected. The exposure time T is determined according to the ambient light intensity E and the camera sensitivity S, satisfying the formula Where β is an empirical coefficient, and the focal length f is determined according to the size of the automobile exhaust device and the shooting distance, so that the automobile exhaust device can be clearly presented in the image while avoiding image distortion.

3. A defect detection method for automobile exhaust device according to claim 1, characterized in that: In step S4, a dynamic standard feature model is constructed, and the extracted feature information is compared with the preset defect-free standard feature using an intelligent comparison algorithm to determine whether the automobile exhaust device has defects. The implementation steps are: S41, firstly collect images and depth information of multiple defect-free automobile exhaust devices, as well as corresponding infrared images, to construct an initial defect-free standard feature set S0; S42, then the dynamic update mechanism is implemented to continuously add the features of the newly detected defect-free samples to the standard feature set, and the standard feature set is optimized and updated regularly. updated =S0∪F new -{F outdated }, where F new is the newly detected defect-free sample feature, {F outdated } are features that have been judged to be outdated and inaccurate; S43. Finally, the similarity between the extracted features of the automobile exhaust device to be detected and the features in the standard feature set is calculated by the mean absolute error from scratches, concave and convex features and heat-related features, and the similarities of multiple dimensions are integrated to obtain an overall similarity index, the formula is S=γ scratch S scratch +γ dent S dent +γ bump S bump +γ heat S heat , if the overall similarity is lower than the threshold value of 0.75, it is judged that the automobile exhaust device has defects; for the automobile exhaust device with defects, the specific location of the defect is determined by analyzing the position information in the extracted features.

Images