An image processing-based automobile part spraying optimization method

Abstract

The present application relates to a kind of based on image processing automobile parts spraying optimization method, to solve the problem such as high missed detection rate in traditional spraying defect detection, inaccurate positioning, parameter optimization lag etc. Method is collected by integrating wide-angle and micro-lens lens spraying robot image, establishes double-lens three-dimensional geometric mapping model, realizes pixel-level alignment and adaptive feature fusion;Combined with material perception texture analysis and edge correction mechanism, improve the recognition accuracy of complex surface defects;Based on edge perception attention network, predict the edge area, guide the micro-lens to carry out adaptive local collection to defect candidate area;Use multiscale segmentation to extract defect features, and input condition diffusion model with current spraying parameter, material code, generate multiple sets of optimized spraying parameters;After simulation evaluation, drive robot to execute compensation control, form closed-loop optimization system.

Description

Technical Field

[0001] This invention relates to the technical field of automotive parts spraying methods, specifically to an image processing-based automotive parts spraying optimization method. Background Technology

[0002] Existing technologies mostly rely on manual visual inspection or traditional image processing algorithms (such as Sobel edge detection), which have limited ability to identify defects in low-contrast, blurred edges, and complex textured surfaces, with a false negative rate as high as 15-20%.

[0003] The global texture captured by the wide-angle lens and the local details acquired by the macro lens are difficult to effectively align and blend in the feature space, resulting in defect localization deviation and affecting the accuracy of subsequent compensation control.

[0004] Traditional parameter optimization methods rely on fixed simulation models, which deviate from the actual spraying process and have high computational complexity, making it difficult to achieve real-time adaptive compensation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an image processing-based method for optimizing the coating of automotive parts. This method significantly improves defect identification accuracy and coating adaptability through dual-lens collaborative imaging and pixel-level alignment, edge-aware guided local detection, and parameter optimization closed-loop control based on a conditional diffusion model, thereby achieving intelligent optimization and quality control of the automotive parts coating process.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing the spraying of automotive parts based on image processing, comprising the following steps. S1. Obtain global images of automotive parts using a painting robot that integrates wide-angle and macro lenses. ; S2. Establish a three-dimensional geometric mapping model between wide-angle and macro lenses to achieve pixel-level alignment between wide-angle and macro images, and perform feature fusion based on adaptive hybrid weights. S2.1, Global Image The denoised global image is obtained by Gaussian filtering. , The global image Includes wide-angle images and macro images The wide-angle image after Gaussian filtering is obtained. and macro images ; S2.2, Denoising the wide-angle image and macro images Feature point extraction was performed on each feature point to obtain its corresponding feature point set. , and the number of features ; S2.3, for the feature point set , Feature matching is performed to obtain a set of matching pairs. and the number of matches ; S2.4, for the feature point set , and matching pair set Geometric transformation estimation yields the homography matrix. Interior point rate set of interior points ; S2.5, Macro Images Homography matrix Image warping and alignment are performed to obtain the warped macro image. and deformation error ; S2.6, Wide-angle image Alignment quality assessment is performed to obtain an alignment quality score. and texture similarity map ; S2.7, The alignment quality score obtained from the evaluation The quality level is determined by quality level assessment. ; S2.8, Based on texture similarity map quality level Wide-angle image macro images Perform fusion weight calculation to obtain the fusion weight map and edge intensity map , ; S2.9, Wide-angle image macro images fusion weight graph Image fusion is performed to obtain the final fused image. ; S3. Predict edge regions of wide-angle images based on an edge-aware attention network, wherein the edge-aware attention network includes an edge prediction branch, an edge-aware attention module, and a feature selection fusion module; S4. Based on the edge prediction results, guide the macro lens to perform adaptive local image acquisition of the defect candidate area. S5. Perform multi-scale defect segmentation and feature extraction on the acquired local images to obtain information on the type, location, size, and severity of spraying defects; S6. Input the defect features, current spraying parameters, and interior part material codes into the conditional diffusion model to generate multiple sets of optimized spraying parameters; S7. Select the optimal parameter combination through simulation evaluation and drive the painting robot to perform compensation control.

[0007] Furthermore, S2.6 also includes: Collect material feature vectors of automotive parts ,in These represent highly reflective materials and complex textured materials, respectively. Based on wide-angle images Macro images Interior point ratio Deformation error Material feature vector To construct material-aware texture analysis: ; in In position Texture similarity at the location, It is the preprocessed wide-angle image in position The pixel vector, It is the deformed macro image at the position The pixel vector, It is the material weighting factor. This is a material compensation item.

[0008] Furthermore, S2.6 also includes reflection compensation and edge processing based on the material. Reflection compensation for highly reflective materials: ; ; in, This is the edge correction value for wide-angle images of highly reflective materials. This represents the edge correction value for macro images of highly reflective materials. The reflectivity of highly reflective materials, The reflection attenuation coefficient is... Compensation is based on the inverse of texture regularity; Edge correction for complex textures: ; in This is the edge correction value for wide-angle images with complex textures. This is the coarse magnification factor. For low reflection compensation; ; in This is the edge correction value for macro images of complex textured materials. To reduce the roughness amplification factor, For low-regularity compensation; Furthermore, S2.7 also includes constructing a material-aware alignment quality score: ; in The weighting coefficients for dynamically allocated internal point rates. For interior point ratio, These are dynamically assigned error weighting coefficients. For deformation error, The maximum permissible error is related to the material. For dynamically assigned texture similarity weights, To show the similarity of average texture. As a weight for material correlation, For material continuity indicators; Construct material quality correction factors and final quality score: ; in The higher the degree of rule fulfillment, the larger its value. The stronger the reflectivity, the smaller its value; The higher the roughness, the larger its value. The lower the reflectivity, the smaller the value. Final quality score: ; Where Q is the preliminary calculated quality score. This is the final alignment quality score after material correction.

[0009] Furthermore, a quality grading system is constructed based on the quality score. The formula for judgment is as follows: ; in These are the thresholds for judging high-quality images and the thresholds for judging quality images, respectively.

[0010] Furthermore, the fusion weights include: Geometric weights: ; in These are the actual observed micro coordinates. For wide-angle image coordinates, For the predicted micro coordinates, These are the geometric weighting smoothing parameters; Texture similarity weight: ; The formula for constructing edge weights is as follows: ; in In position Edge consistency weights at the location, It is a wide-angle edge corrected according to the material type. These are macro edges corrected according to material type. It is a small parameter to prevent division by zero; The formula for weighted combination is as follows: ; in, It is with quality grading Changing adjustment parameters; ; Fusion Weight Graph .

[0011] Furthermore, the adaptive weighted fusion of images in S2.9 is specifically as follows: Input the preprocessed wide-angle image, the deformed and aligned macro image, the final alignment quality score, and the weighted fusion formula constructed from the material feature vector: ; in For adaptive weighting function: ; in The bias term is used to control the weight distribution. The mass fraction influence coefficient. The texture regularity influence coefficient. This is the roughness influence coefficient.

[0012] Compared with the prior art, the technical solution of this application has the following beneficial effects: I. This image processing-based optimization method for automotive component painting employs a dual-lens approach combining wide-angle and macro lenses for collaborative acquisition. It utilizes a 3D geometric mapping model to achieve pixel-level image alignment and adaptive feature fusion, effectively overcoming the limitations of traditional methods in identifying defects on low-contrast, blurred edges, and complex textured surfaces. By combining material-aware texture analysis and edge correction mechanisms, the method significantly improves the accuracy and robustness of defect detection, potentially reducing the false negative rate from the current 15-20%. II. This image processing-based optimization method for automotive component painting, by introducing an edge-aware attention network, enables the system to intelligently predict edge regions in images and guide the macro lens to adaptively acquire candidate defect regions. This method not only improves the targeting of detection but also reduces invalid scans, enhances system response speed and resource utilization efficiency, and provides high-quality input for subsequent defect segmentation and feature extraction.

[0013] III. This image processing-based method for optimizing automotive component painting inputs extracted defect features, current painting parameters, and material codes into a conditional diffusion model to generate multiple sets of optimized painting parameters. The optimal combination is then selected through simulation evaluation to drive a robot to perform compensatory painting. The entire process forms a closed-loop control system that dynamically adjusts the painting strategy based on real-time detection results, achieving continuous optimization and defect rate control in the painting process. It possesses strong engineering applicability and real-time adaptive capabilities. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0016] This embodiment provides an image processing-based optimization method for automotive component painting. First, a wide-angle lens and a macro lens are integrated and installed at the end effector of a painting robot arm. The wide-angle lens is used to acquire a global image of the entire area of ​​the automotive component, while the macro lens is used for high-resolution imaging of local areas. The painting robot is then activated and moves along a preset path, simultaneously triggering the dual lenses to acquire images.

[0017] A 3D geometric mapping model between a wide-angle lens and a macro lens is established. By jointly calibrating the two cameras, the rotation matrix and translation vector between them are obtained, achieving pixel-level spatial position mapping. Gaussian filtering is applied to both the acquired wide-angle and macro images to remove sensor noise. SIFT or ORB feature points are extracted from the two images, and feature matching is performed to obtain matching point pairs. The homography matrix is ​​estimated based on the RANSAC algorithm, and the macro image is deformed to the coordinate system of the wide-angle image to achieve spatial alignment.

[0018] Alignment quality is evaluated based on metrics such as inlier rate, deformation error, and texture similarity to obtain an alignment quality score, which is then used to determine the quality level. Combining the texture similarity map and the quality level, a fusion weight map is calculated, and adaptive weighted fusion is performed on the wide-angle image and the deformed macro image to obtain a fused image that simultaneously contains global structural information and local detail information.

[0019] The fused image is input into a pre-trained edge-aware attention network, which includes an edge prediction branch and outputs an edge probability map. The edge-aware attention module assigns higher weights to edge regions. The feature selection fusion module selectively fuses multi-level features to generate edge region prediction results.

[0020] Based on the edge prediction results, the system automatically marks candidate defect regions and guides the macro lens to the corresponding position for secondary high-resolution local acquisition. A multi-scale segmentation algorithm is then used on the acquired local images to extract the precise contours of the defects and calculate the defect type, location coordinates, size, and severity level.

[0021] The aforementioned defect characteristics, parameters such as pressure, flow rate, and atomizing air pressure of the current spraying equipment, as well as the component material coding information, are input into the conditional diffusion model. The model undergoes a multi-step iterative denoising process to generate multiple sets of optimized spraying parameter schemes. Each scheme is evaluated using spraying simulation software through virtual spraying, comparing indicators such as defect coverage and film thickness uniformity to select the optimal parameter combination. Finally, the optimal parameters are sent to the spraying robot controller to execute the compensatory spraying operation, forming a closed-loop control process of detection-optimization-execution.

[0022] Before the actual spraying operation, the material type of the part to be sprayed is input via the host computer, and the system automatically retrieves the corresponding material feature vector. During the alignment quality assessment in step S2.6, this material feature vector is read synchronously.

[0023] For each pixel location in the image, local pixel vectors are extracted from the wide-angle image and the distorted macro image, and the similarity between them is calculated. A material weighting factor and a material compensation term are introduced. When a highly reflective material is detected, the weighting factor automatically increases the tolerance for highlight areas; when a complex texture material is detected, the compensation term enhances the sensitivity to texture details. Through material-aware texture similarity calculation, a pixel-by-pixel texture similarity map is obtained. This map can realistically reflect the true alignment degree of the dual-lens images under different material surfaces, providing an accurate basis for subsequent weight fusion.

[0024] When the system identifies the current component as being made of a highly reflective material, both the wide-angle and macro images show areas of strong specular reflection, causing edge information to be obscured by highlights. In this case, after extracting edge features from the wide-angle image, nonlinear suppression of edge intensity is applied based on the reflectivity and attenuation coefficients. Simultaneously, compensation is provided for areas with low texture regularity, allowing the faint edges that were originally obscured by highlights to become visible. A similar compensation mechanism is used in the macro image to ensure consistency of edge features in highly reflective areas between the two lenses.

[0025] When the system identifies the current component as having a complex texture, the surface has dense texture lines that can easily be confused with the edges of real defects. In this case, the edges of the wide-angle image are moderately roughened to enhance the contrast between the defect edges and the background texture; for macro images, a reduced roughness magnification factor is used to avoid over-enhancing and causing textures to be misjudged as defects. At the same time, compensation is applied to low-regularity areas to make the edge extraction of messy texture areas more robust.

[0026] After material-adaptive edge correction, corrected wide-angle edge maps and macro edge maps are obtained, providing high-quality edge-consistent input for subsequent fusion weight calculation.

[0027] The inlier rate weighting coefficient and error weighting coefficient adopt a dynamic allocation strategy: when the inlier rate is high, the system considers the geometric transformation estimation reliable and appropriately reduces the error weight; when the inlier rate is low, the error weight is increased to reflect alignment uncertainty. Deformation error is normalized to the maximum permissible error range related to the material. This maximum permissible error is preset based on material characteristics; highly reflective materials are allowed larger errors, while materials with fine textures require smaller errors. Texture similarity is taken as the average of the entire image and jointly evaluated with the material continuity index.

[0028] After initially calculating the quality score Q, a material quality correction factor is introduced. For materials with high regularity, the correction factor is positively amplified; for materials with high reflectivity, the correction factor is negatively suppressed; for materials with high roughness, the correction factor is appropriately increased; and for materials with low reflectivity, the correction factor is appropriately decreased. Finally, a material-corrected alignment quality score is obtained, which accurately reflects the actual usability of dual-lens image alignment on different material surfaces.

[0029] When the final alignment quality score is greater than or equal to the high quality threshold, it is judged as high quality, indicating that the alignment accuracy of the dual-lens images is extremely high and the quality of the fused image is excellent, which can be directly used for subsequent defect detection; when the quality score is between the qualified threshold and the high quality threshold, it is judged as qualified, indicating that the alignment quality meets the basic requirements and the fused image can be used for detection but the confidence level is slightly low; when the quality score is lower than the qualified threshold, it is judged as low quality, the system automatically alarms and triggers re-acquisition to avoid missed or false defects due to image alignment failure.

[0030] The quality level determination result is synchronously transmitted to the subsequent weight calculation module for dynamic adjustment of the fusion strategy.

[0031] Geometric weights are calculated based on the deviation between the actual observed coordinates and the predicted coordinates of the macro image; the smaller the deviation, the higher the weight, reflecting the local reliability of spatial alignment. Texture similarity weights are directly mapped from the texture similarity map generated in step S2.6; the more similar the textures, the higher the weight. Edge consistency weights are calculated based on the corrected wide-angle edge map and the macro edge map; the closer the intensity of the two edge maps at the same location, the higher the weight.

[0032] The three weights are fused using a weighted combination method, with the combination coefficients adaptively adjusted according to the quality level. When the quality level is high, the contributions of geometric and edge weights are increased; when the quality level is acceptable, the proportion of texture similarity weight is appropriately increased to compensate for insufficient alignment accuracy. Finally, a pixel-by-pixel fusion weight map of the same size as the image is generated to guide subsequent image fusion.

[0033] An adaptive weighting function is constructed, which includes a bias term, a quality score influence coefficient, a texture regularity influence coefficient, and a roughness influence coefficient. The bias term controls the basic weight allocation between the wide-angle image and the macro image during fusion; the quality score influence coefficient increases the contribution weight of the macro image when the alignment quality is higher; the texture regularity influence coefficient gives higher weight to the macro image on surfaces with regular textures; and the roughness influence coefficient moderately suppresses the weight of the macro image on rough surfaces to avoid over-enhancing the texture.

[0034] During fusion, for each pixel location, the fusion coefficients of the wide-angle image and the macro image are calculated according to the aforementioned adaptive weighting function, and then weighted and summed to obtain the final fused image. This fused image preserves the global structural integrity of the wide-angle image while incorporating the local detail information of the macro image. Furthermore, it can adaptively adjust the fusion strategy under different material and alignment quality conditions, providing high-quality input for subsequent edge prediction and defect segmentation.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the spraying of automotive parts based on image processing, comprising the following steps, characterized in that: S1. Obtain global images of automotive parts using a painting robot that integrates wide-angle and macro lenses. ; S2. Establish a three-dimensional geometric mapping model between wide-angle and macro lenses to achieve pixel-level alignment between wide-angle and macro images, and perform feature fusion based on adaptive hybrid weights. S2.1, Global Image The denoised global image is obtained by Gaussian filtering. , The global image Includes wide-angle images and macro images After Gaussian filtering, the denoised wide-angle image is obtained. and macro images ; S2.2, Denoising the wide-angle image and macro images Feature point extraction was performed on each feature point to obtain its corresponding feature point set. , and the number of features ; S2.3, for the feature point set , Feature matching is performed to obtain a set of matching pairs. and the number of matches ; S2.4, for the feature point set , and matching pair set Geometric transformation estimation yields the homography matrix. Interior point rate set of interior points ; S2.5, Macro Images Homography matrix Image warping and alignment are performed to obtain the warped macro image. and deformation error ; S2.6, Wide-angle image Alignment quality assessment is performed to obtain an alignment quality score. and texture similarity map ; S2.7, The alignment quality score obtained from the evaluation The quality level is obtained by determining the quality level. ; S2.8, Based on texture similarity map quality level Wide-angle image macro images Perform fusion weight calculation to obtain the fusion weight map and edge intensity map , ; S2.9, Wide-angle image macro images fusion weight graph Image fusion is performed to obtain the final fused image. ; S3. Predict edge regions of wide-angle images based on an edge-aware attention network, wherein the edge-aware attention network includes an edge prediction branch, an edge-aware attention module, and a feature selection fusion module; S4. Based on the edge prediction results, guide the macro lens to perform adaptive local image acquisition of the defect candidate area. S5. Perform multi-scale defect segmentation and feature extraction on the acquired local images to obtain information on the type, location, size, and severity of spraying defects; S6. Input the defect features, current spraying parameters, and interior part material codes into the conditional diffusion model to generate multiple sets of optimized spraying parameters; S7. Select the optimal parameter combination through simulation evaluation and drive the painting robot to perform compensation control.

2. The image processing-based method for optimizing automotive component coating as described in claim 1, characterized in that: S2.6 also includes: Collect material feature vectors of automotive parts ,in These represent highly reflective materials and complex textured materials, respectively. Based on wide-angle images Macro images Interior point ratio Deformation error Material feature vector To construct material-aware texture analysis: ； in In position Texture similarity at the location, It is the preprocessed wide-angle image in position The pixel vector, It is the deformed macro image at the position The pixel vector, It is the material weighting factor. This is a material compensation item.

3. The image processing-based method for optimizing automotive component coating as described in claim 1, characterized in that: S2.6 also includes reflection compensation and edge processing based on the material; Reflection compensation for highly reflective materials: ； ； in, This is the edge correction value for wide-angle images of highly reflective materials. This represents the edge correction value for macro images of highly reflective materials. The reflectivity of highly reflective materials, The reflection attenuation coefficient is... Compensation is based on the inverse of texture regularity; Edge correction for complex textures: ； in This is the edge correction value for wide-angle images with complex textures. This is the coarse magnification factor. For low reflection compensation; ； in This is the edge correction value for macro images of complex textured materials. To reduce the roughness amplification factor, This is for low-regularity compensation.

4. The image processing-based method for optimizing automotive component coating as described in claim 1, characterized in that: S2.7 also includes constructing a material-aware alignment quality score: ； in The weighting coefficients for dynamically allocated internal point rates. For interior point ratio, These are dynamically assigned error weighting coefficients. For deformation error, The maximum permissible error is related to the material. For dynamically assigned texture similarity weights, To show the similarity of average texture. As a weight for material correlation, For material continuity indicators; Construct material quality correction factors and final quality score: ； in The higher the degree of rule fulfillment, the larger its value. The stronger the reflectivity, the smaller its value; The higher the roughness, the larger its value. The lower the reflectivity, the smaller the value. Final quality score: ； Where Q is the preliminary calculated quality score. This is the final alignment quality score after material correction.

5. The image processing-based method for optimizing automotive component painting according to claim 1, characterized in that: Quality grading is constructed based on quality scores. The formula for judgment is as follows: ； in, These are the thresholds for judging high-quality images and the thresholds for judging quality images, respectively.

6. The image processing-based method for optimizing automotive component coating as described in claim 1, characterized in that: The fusion weights include: Geometric weights: ； in, These are the actual observed micro coordinates. For wide-angle image coordinates, For the predicted micro coordinates, These are the geometric weighting smoothing parameters; Texture similarity weight: ； The formula for constructing edge weights is as follows: ； in In position Edge consistency weight, It is a wide-angle edge corrected according to the material type. These are macro edges corrected according to material type. It is a small parameter to prevent division by zero; The formula for weighted combination is as follows: ； in, It is with quality grading Changing adjustment parameters ； Fusion Weight Graph .

7. The image processing-based method for optimizing automotive component painting according to claim 1, characterized in that: The adaptive weighted fusion of images in S2.9 is as follows: Input the preprocessed wide-angle image, the deformed and aligned macro image, the final alignment quality score, and the weighted fusion formula constructed from the material feature vector: ； in For adaptive weighting function: ； in The bias term is used to control the weight distribution. The mass fraction influence coefficient. The texture regularity influence coefficient. This is the roughness influence coefficient.

Images