Automobile part detection method

Abstract

The invention relates to the technical field of automobile part detection, in particular to an automobile part detection method, which comprises the following steps of: dividing a region range of a to-be-detected surface of an automobile part to obtain a plurality of image acquisition regions, and setting acquisition points for the plurality of image acquisition regions; acquiring basic feature data of a plurality of image acquisition areas, setting corresponding acquisition environment light sources, and performing image acquisition by an image acquisition camera corresponding to the acquisition environment light sources to obtain a plurality of detection images; performing feature analysis comprehensive processing on the plurality of detection images to obtain a defect area candidate image, and performing defect feature enhancement processing on the defect area candidate image to obtain a defect positioning image; and inputting the defect positioning image into the pre-training model, and outputting a detection result according to the defect parameters. According to the invention, the acquisition effect of the detection image is guaranteed, the visibility of the detection image is improved, and the accuracy of defect identification is guaranteed while the detection efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of automotive parts testing technology, and more specifically, to a method for testing automotive parts. Background Technology

[0002] Automotive parts are an important category of automotive components. As the foundation of a car's overall performance and safety, there are many types of automotive parts. To ensure the reliability of a car during use, it is necessary to detect defects in automotive parts.

[0003] In current methods for surface defect detection of automotive parts, defect analysis is often performed by acquiring an overall image of the part's surface. However, in practical applications, the overall image acquired is subject to low visibility due to the material and irregular surface of the automotive parts, as well as the influence of the acquisition location and ambient light, which in turn affects the accuracy of defect identification.

[0004] Based on this, a method for testing automotive parts is proposed. Summary of the Invention

[0005] The main objective of this invention is to provide a method for testing automotive parts to overcome the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides a method for testing automotive parts, comprising:

[0007] The process involves obtaining the area of ​​the automotive component to be inspected, obtaining the visual shooting range of the image acquisition camera, setting a reference bounding box based on the visual shooting range, segmenting the area of ​​the automotive component to be inspected based on the reference bounding box, obtaining several image acquisition areas, selecting the center points of several image acquisition areas to make vertical auxiliary lines, and setting the other end of the auxiliary lines as the acquisition position of the acquisition camera.

[0008] Acquire basic feature data of several image acquisition areas of automotive parts, set several sets of acquisition ambient light sources based on the basic feature data, and have the image acquisition camera acquire a set of images corresponding to each set of acquisition ambient light sources to obtain several detection images.

[0009] Feature analysis and comprehensive processing are performed on several detection images to obtain candidate images of defect regions. Defect feature enhancement processing is then performed on the candidate images of defect regions to obtain defect location images.

[0010] Image synthesis technology is used to generate defect samples to expand the dataset using a diffusion model. The dataset is then input into a pre-trained model for defect detection training. The defect training parameters of the pre-trained model are reused to input defect location images into the pre-trained model, and the detection results are output after adapting the defect parameters.

[0011] As a further improvement of the present invention, the specific steps for setting the acquisition position of the acquisition camera include:

[0012] Extract the closed contour line of the surface to be inspected on the automotive parts, and select the area within the closed contour line to mark the area to be inspected;

[0013] Set the sensor dimensions of the image acquisition camera to width w, height h, focal length f, and working distance WD, and substitute them into the formula. Obtain the actual field of view SH, extract the contour closure line of the actual field of view, and select the area within the contour closure line as the visualization shooting area;

[0014] A baseline bounding box is set based on the closed contour line of the visualized shooting area. The baseline bounding box is then aligned with the area to be detected using any closed contour line as a parallel line to complete the selection and segmentation of the area to be detected, resulting in several image acquisition areas. The center points of several image acquisition areas are selected to create vertical auxiliary lines, and the length of the vertical auxiliary lines is equal to the working distance WD of the image acquisition camera.

[0015] As a further improvement of the present invention, the specific steps for acquiring several detection images include:

[0016] The basic feature data of the image acquisition area of ​​automotive parts include material and surface regularity;

[0017] Based on the fact that automotive parts have a regular, highly reflective metallic surface, the corresponding ambient light sources for data acquisition are set up as follows: bright field illumination, diffuse coaxial light source, blue light source with a visible light band of 450-495nm; dark field illumination, diffuse coaxial light source, blue light source with a visible light band of 450-495nm.

[0018] Based on the fact that automotive parts are irregular surfaces with high reflectivity, the corresponding ambient light sources for data collection include bright field illumination, diffuse backlight, blue light source with visible light band of 450-495nm, and dark field illumination, diffuse backlight, blue light source with visible light band of 450-495nm.

[0019] The image acquisition camera sequentially acquires images corresponding to the ambient light source, resulting in several detection images.

[0020] As a further improvement of the present invention, the specific steps for feature analysis and comprehensive processing of several detected images include:

[0021] The image resolution of the detected image is divided into several pixel blocks. Feature quantization is performed on the pixel blocks to extract image features, including texture, color difference and shape. The sliding detection window covers the entire image and performs image feature connectivity detection on adjacent pixel blocks. Based on the connectivity detection, discontinuous pixel blocks are marked as suspected defect areas.

[0022] Gray-level variance processing was performed on pixel blocks in suspected defect areas. OpenCV was used to read the image and convert it to a single-channel grayscale image to obtain grayscale data. Based on the grayscale data, the formula was used... The grayscale variance FC is obtained, where μ is the mean grayscale value, N is the total number of pixels in the image, and X is the mean grayscale value. i The grayscale value for each pixel;

[0023] Set the standard deviation σ, and substitute the gray mean μ of the pixel block into the formula FC'=μ+k×σ to obtain the dynamic variance threshold, where k is an empirical coefficient, usually a constant of 1-3, used to balance noise suppression and defect sensitivity. Compare the gray variance value FC with the variance threshold FC'. If FC>FC', a high variance region is set, and the suspected defect region contains defect mutations. The current detection image is marked as a candidate image for the defect region. If FC<FC', a low variance region is set, and the suspected defect region corresponds to a uniform background block, and defects are quickly eliminated.

[0024] The candidate image of the defect area is input into the gamma and linear blending model. Gamma correction is applied to low grayscale pixel blocks, and linear stretching is applied to high grayscale blocks. For example, if the gamma value γ < 1, the dark details are stretched and the bright parts are compressed, which is suitable for images with overly dark dark areas; if γ > 1, the dark areas are compressed and the bright parts are enhanced, which is suitable for images with overexposed highlights; if γ = 1, there is no adjustment and the linear relationship is maintained.

[0025] Using formula The output grayscale value s is obtained, where r is the input grayscale value, a controls the contrast, b adjusts the brightness, and c is the brightness scaling factor, which is a constant ranging from 1 to 255. i ' is the grayscale threshold. Based on the output grayscale, the dark and bright details of the candidate image pixels in the defect area are adjusted to reconstruct the defect area image, highlighting the defect outline, and thus obtaining the defect localization image.

[0026] As a further improvement of the present invention, the specific improvements to the defect parameter output detection results include:

[0027] The noise figure β during the diffusion process is controlled by forward and reverse algorithms of the diffusion model using the input defect sample image. t The strength control module SDAS is introduced to control the intensity of defect generation;

[0028] For example, setting β max =0.02 generates microcracks, β max =0.1 produces obvious corrosion marks;

[0029] Multimodal defect generation involves classifying input defect sample images into metal, plastic, and glass materials. A pre-trained LoRA adapter is applied to each material, and the diffusion model is adjusted to generate defect shapes that conform to physical properties. An ambient light intensity parameter L∈[0,1] is added to the defect shape, and the rendering equation I is used.gen =I originaI • (1+L·Δ) corrects the light reflection characteristics of the generated defects;

[0030] The defect morphology and intensity generated by the diffusion model are input into the Ray framework to generate defect example images in parallel. The defect example images are combined into a dataset and input into the pre-trained model for defect detection training. The output layer of the pre-trained model integrates a multi-task decoder to generate defect categories, location coordinates, and morphological parameters, such as crack length and pore diameter. The defect location image is input into the pre-trained model to adapt the defect parameters and output the detection results.

[0031] The beneficial effects of this invention are:

[0032] This invention analyzes the acquisition position of the acquisition camera and the basic features of automotive parts to set the acquisition points and ambient light sources for the corresponding detection images. This avoids the problem of low visibility of the acquired detection images caused by the influence of automotive parts and the acquisition environment in practical applications. It helps to ensure the acquisition effect of the detection images, improve the visibility of the detection images, and further improve the accuracy of defect identification.

[0033] This invention performs comprehensive processing on the detected image, enhances and highlights the defective region of the defective image, and completes the detection of the defective image by training with defective examples of the pre-trained model. This helps to improve the detection efficiency while ensuring the accuracy of defect identification. Attached Figure Description

[0034] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0035] Figure 1 This is a flowchart of the present invention;

[0036] Figure 2 This is the image acquisition area division diagram of the present invention. Detailed Implementation

[0037] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0038] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0039] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of the invention described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0040] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0041] Please see Figures 1-2 As shown, a method for testing automotive parts includes:

[0042] The process involves obtaining the area of ​​the automotive component to be inspected, obtaining the visual shooting range of the image acquisition camera, setting a reference bounding box based on the visual shooting range, segmenting the area of ​​the automotive component to be inspected based on the reference bounding box, obtaining several image acquisition areas, selecting the center points of several image acquisition areas to make vertical auxiliary lines, and setting the other end of the auxiliary lines as the acquisition position of the acquisition camera.

[0043] The specific steps for setting the acquisition camera's acquisition location include:

[0044] Extract the closed contour line of the surface to be inspected on the automotive parts, and select the area within the closed contour line to mark the area to be inspected;

[0045] Set the sensor dimensions of the image acquisition camera to width w, height h, focal length f, and working distance WD, and substitute them into the formula. Obtain the actual field of view SH, extract the contour closure line of the actual field of view, and select the area within the contour closure line as the visualization shooting area;

[0046] Based on the contour closure line of the visualized shooting area, a reference bounding box is set. The reference bounding box and the area to be detected are aligned with any contour closure line as a parallel line to complete the selection and segmentation of the area to be detected, resulting in several image acquisition areas. The center points of several image acquisition areas are selected to make vertical auxiliary lines. The length of the vertical auxiliary lines is equal to the working distance WD of the image acquisition camera.

[0047] It should be noted that, see Figure 2 When several image acquisition areas are obtained by coinciding the reference bounding box and the region to be detected with any contour closure line as a parallel line, the preferred choice of any contour closure line should be that there is no gap between the image acquisition areas during the coinciding process.

[0048] The working distance WD of an image acquisition camera is the physical distance from the camera lens to the surface of the image acquisition area.

[0049] Acquire basic feature data of several image acquisition areas of automotive parts, set several sets of acquisition ambient light sources based on the basic feature data, and have the image acquisition camera acquire a set of images corresponding to each set of acquisition ambient light sources to obtain several detection images.

[0050] The specific steps for acquiring several detection images include:

[0051] The basic feature data of the image acquisition area of ​​automotive parts include material and surface regularity;

[0052] Based on the fact that automotive parts have a regular, highly reflective metallic surface, the corresponding ambient light sources for data acquisition are set up as follows: bright field illumination, diffuse coaxial light source, blue light source with a visible light band of 450-495nm; dark field illumination, diffuse coaxial light source, blue light source with a visible light band of 450-495nm.

[0053] Based on the fact that automotive parts are irregular surfaces with high reflectivity, the corresponding ambient light sources for data collection include bright field illumination, diffuse backlight, blue light source with visible light band of 450-495nm, and dark field illumination, diffuse backlight, blue light source with visible light band of 450-495nm.

[0054] The image acquisition camera sequentially acquires images corresponding to the ambient light source, resulting in several detection images.

[0055] It should be noted that the bright field illumination source is located within the reflective cone of the camera's field of view, and the light illuminates the surface of the acquisition area at a high angle or a perpendicular angle.

[0056] Dark field illumination sources are used to form an inverted hollow light cone through special condenser lenses, such as a cardioid reflecting condenser lens, so that the incident light is obliquely incident on the surface of the acquisition area at an extremely low angle.

[0057] The diffuse reflection coaxial light source uses a beam splitter to make the light source and imaging optical path coaxial, and uses a diffuse reflection plate to eliminate stray light interference, capturing only the specular reflection signal of the object surface to achieve high contrast imaging. The light source, beam splitter and acquisition camera optical axis are completely coaxial, avoiding interference from multiple angles of light.

[0058] The diffuse backlight source evenly scatters the light emitted by the LED array through the diffuser plate, forming a large area of ​​soft illumination. It eliminates the directivity of direct light through diffuse reflection, allowing the light to penetrate the object being measured evenly and highlighting the contrast differences of the surface structure.

[0059] Feature analysis and comprehensive processing are performed on several detection images to obtain candidate images of defect regions. Defect feature enhancement processing is then performed on the candidate images of defect regions to obtain defect location images.

[0060] The specific steps for feature analysis and comprehensive processing of several detected images include:

[0061] The image resolution of the detected image is divided into several pixel blocks. Feature quantization is performed on the pixel blocks to extract image features, including texture, color difference and shape. The sliding detection window covers the entire image and performs image feature connectivity detection on adjacent pixel blocks. Based on the connectivity detection, discontinuous pixel blocks are marked as suspected defect areas.

[0062] Gray-level variance processing was performed on pixel blocks in suspected defect areas. OpenCV was used to read the image and convert it to a single-channel grayscale image to obtain grayscale data. Based on the grayscale data, the formula was used... The grayscale variance FC is obtained, where μ is the mean grayscale value, N is the total number of pixels in the image, and X is the mean grayscale value. i The grayscale value for each pixel;

[0063] Set the standard deviation σ, and substitute the gray mean μ of the pixel block into the formula FC'=μ+k×σ to obtain the dynamic variance threshold, where k is an empirical coefficient, usually a constant of 1-3, used to balance noise suppression and defect sensitivity. Compare the gray variance value FC with the variance threshold FC'. If FC>FC', a high variance region is set, and the suspected defect region contains defect mutations. The current detection image is marked as a candidate image for the defect region. If FC<FC', a low variance region is set, and the suspected defect region corresponds to a uniform background block, and defects are quickly eliminated.

[0064] The candidate image of the defect area is input into the gamma and linear blending model. Gamma correction is applied to low grayscale pixel blocks, and linear stretching is applied to high grayscale blocks. For example, if the gamma value γ < 1, the dark details are stretched and the bright parts are compressed, which is suitable for images with overly dark dark areas; if γ > 1, the dark areas are compressed and the bright parts are enhanced, which is suitable for images with overexposed highlights; if γ = 1, there is no adjustment and the linear relationship is maintained.

[0065] Using formula The output grayscale value s is obtained, where r is the input grayscale value, a controls the contrast, b adjusts the brightness, and c is the brightness scaling factor, which is a constant ranging from 1 to 255. i ' is the grayscale threshold. Based on the output grayscale, the dark and bright details of the candidate image pixels in the defect area are adjusted to reconstruct the defect area image, highlighting the defect outline, and thus obtaining the defect localization image.

[0066] Image synthesis technology is used to generate defect samples to expand the dataset using a diffusion model. The dataset is then input into a pre-trained model for defect detection training. The defect training parameters of the pre-trained model are reused to input defect location images into the pre-trained model, and the detection results are output after adapting the defect parameters.

[0067] The specific output detection results for the adapted defect parameters include:

[0068] The noise figure β during the diffusion process is controlled by forward and reverse algorithms of the diffusion model using the input defect sample image. t The strength control module SDAS is introduced to control the intensity of defect generation;

[0069] For example, setting β max =0.02 generates microcracks, β max =0.1 produces obvious corrosion marks;

[0070] Multimodal defect generation involves classifying input defect sample images into metal, plastic, and glass materials. A pre-trained LoRA adapter is applied to each material, and the diffusion model is adjusted to generate defect shapes that conform to physical properties. An ambient light intensity parameter L∈[0,1] is added to the defect shape, and the rendering equation I is used. gen =I originaI • (1+L·Δ) corrects the light reflection characteristics of the generated defects;

[0071] The defect morphology and intensity generated by the diffusion model are input into the Ray framework to generate defect example images in parallel. The defect example images are combined into a dataset and input into the pre-trained model for defect detection training. The output layer of the pre-trained model integrates a multi-task decoder to generate defect categories, location coordinates, and morphological parameters, such as crack length and pore diameter. The defect location image is input into the pre-trained model to adapt the defect parameters and output the detection results.

[0072] The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for testing automotive parts, characterized in that, include: The process involves obtaining the area of ​​the automotive component's surface to be inspected, obtaining the visual shooting range of the image acquisition camera, setting a reference bounding box based on the visual shooting range, segmenting the area of ​​the automotive component's surface to be inspected based on the reference bounding box, obtaining several image acquisition areas, selecting the center points of several image acquisition areas to make vertical auxiliary lines, and setting the other end of the auxiliary lines as the acquisition position of the acquisition camera. Acquire basic feature data of several image acquisition areas of automotive parts, set several sets of acquisition ambient light sources based on the basic feature data, and have the image acquisition camera acquire a set of images corresponding to each set of acquisition ambient light sources to obtain several detection images. Feature analysis and comprehensive processing are performed on several detection images to obtain candidate images of defect regions. Defect feature enhancement processing is then performed on the candidate images of defect regions to obtain defect location images. Image synthesis technology is used to generate defect samples to expand the dataset using a diffusion model. The dataset is then input into a pre-trained model for defect detection training. The defect training parameters of the pre-trained model are reused to input defect location images into the pre-trained model, and the detection results are output after adapting the defect parameters.

2. The method for testing automotive parts according to claim 1, characterized in that, The specific steps for setting the acquisition camera's acquisition location include: Extract the closed contour line of the surface to be inspected on the automotive parts, and select the area within the closed contour line to mark the area to be inspected; Set the sensor dimensions of the image acquisition camera to width w, height h, focal length f, and working distance WD, and substitute them into the formula. Obtain the actual field of view SH, extract the contour closure line of the actual field of view, and select the area within the contour closure line as the visualization shooting area; A baseline bounding box is set based on the closed contour line of the visualized shooting area. The baseline bounding box is then aligned with the area to be detected using any closed contour line as a parallel line to complete the selection and segmentation of the area to be detected, resulting in several image acquisition areas. The center points of several image acquisition areas are selected to create vertical auxiliary lines, and the length of the vertical auxiliary lines is equal to the working distance WD of the image acquisition camera.

3. The method for testing automotive parts according to claim 1, characterized in that, The specific steps for acquiring several detection images include: The basic feature data of the image acquisition area of ​​automotive parts include material and surface regularity; Based on the fact that automotive parts have a regular, highly reflective metallic surface, the corresponding ambient light sources for data acquisition are set up as follows: bright field illumination, diffuse coaxial light source, blue light source with a visible light band of 450-495nm; dark field illumination, diffuse coaxial light source, blue light source with a visible light band of 450-495nm. Based on the fact that automotive parts are irregular surfaces with high reflectivity, the corresponding ambient light sources for data collection include bright field illumination, diffuse backlight, blue light source with visible light band of 450-495nm, and dark field illumination, diffuse backlight, blue light source with visible light band of 450-495nm. The image acquisition camera sequentially acquires images corresponding to the ambient light source, resulting in several detection images.

4. The method for testing automotive parts according to claim 1, characterized in that, The specific steps for feature analysis and comprehensive processing of several detected images include: The image resolution of the detected image is divided into several pixel blocks. Feature quantization is performed on the pixel blocks to extract image features, including texture, color difference and shape. The sliding detection window covers the entire image and performs image feature connectivity detection on adjacent pixel blocks. Based on the connectivity detection, discontinuous pixel blocks are marked as suspected defect areas. Gray-level variance processing was performed on pixel blocks in suspected defect areas. OpenCV was used to read the image and convert it to a single-channel grayscale image to obtain grayscale data. Based on the grayscale data, the formula was used... The grayscale variance FC is obtained, where μ is the mean grayscale value, N is the total number of pixels in the image, and X is the mean grayscale value. i The grayscale value for each pixel; Set the standard deviation σ, and substitute the gray mean μ of the pixel block into the formula FC'=μ+k×σ to obtain the dynamic variance threshold, where k is an empirical coefficient. Compare the gray variance value FC with the variance threshold FC'. If FC>FC', a high variance region is set, and the suspected defect region contains defect mutations. The current detection image is marked as a candidate image for the defect region. If FC<FC', a low variance region is set, and the suspected defect region corresponds to a uniform background block, and defects are quickly eliminated. The candidate image of the defect region is input into a gamma-linear mixture model. Gamma correction is applied to low-grayscale pixel blocks, and linear stretching is applied to high-grayscale blocks. The formula is then used... The output grayscale value s is obtained, where r is the input grayscale value, a controls the contrast, b adjusts the brightness, c is the brightness scaling factor, and X is the output grayscale value. i ' is the grayscale threshold. Based on the output grayscale, the dark and bright details of the candidate image pixels in the defect area are adjusted to reconstruct the defect area image, highlighting the defect outline, and thus obtaining the defect localization image.

5. The method for testing automotive parts according to claim 1, characterized in that, The specific output detection results for the adapted defect parameters include: The noise figure β during the diffusion process is controlled by forward and reverse algorithms of the diffusion model using the input defect sample image. t The strength control module SDAS is introduced to control the intensity of defect generation; Multimodal defect generation involves classifying input defect sample images into metal, plastic, and glass materials. A pre-trained LoRA adapter is applied to each material, and the diffusion model is adjusted to generate defect shapes that conform to physical properties. An ambient light intensity parameter L∈[0,1] is added to the defect shape, and the rendering equation I is used. gen =I originaI • (1+L·Δ) corrects the light reflection characteristics of the generated defects; The defect morphology and intensity generated by the diffusion model are input into the Ray framework to generate defect example images in parallel. The defect example images are combined into a dataset and input into the pre-trained model for defect detection training. The output layer of the pre-trained model integrates a multi-task decoder to generate defect category, location coordinates and morphological parameters. The defect localization image is input into the pre-trained model to adapt the defect parameters and output the detection results.

Images