Vehicle chassis rivet detection method

Abstract

The invention provides a vehicle chassis rivet detection method. The method comprises the following steps: S1, obtaining a vehicle chassis image data set; s2, improving a YOLO v9 neural network according to the characteristics of the chassis rivet image data, and constructing a backbone network, a multi-path attention mechanism, a residual fusion module and an output module; s3, inputting the processed chassis rivet image data into the neural network for training, and obtaining an optimal neural network detection model through back propagation; s4, obtaining a detection position coordinate of the rivet by using the neural network detection model; and S5, after the detection position coordinates of the rivets are compared with the set installation position coordinate data, a detection result that the rivets are installed in place or not installed in place is generated. According to the invention, the detection precision of the vehicle chassis rivet can be well improved, errors caused by manual detection in the prior art are avoided, and the accuracy and safety of rivet installation are ensured.

Description

Technical Field

[0001] This invention belongs to the field of vehicle inspection technology, and specifically relates to a method for inspecting rivets on a vehicle chassis. Background Technology

[0002] Traditional methods for inspecting rivets on vehicle chassis typically rely on manual observation and recording, with operators determining rivet installation status by visual inspection. This traditional method is not only significantly inefficient but also susceptible to human error, leading to missed rivets, misjudgments, or inconsistent results. Operators may experience fatigue due to prolonged work, affecting the accuracy of their judgments. Furthermore, factors such as rivet size, position, and angle can make manual identification difficult, further increasing inspection errors and uncertainties.

[0003] With the increasing demands for production efficiency and quality in the modern automotive industry, traditional manual inspection methods are gradually revealing their shortcomings. This is especially true in complex and dynamic industrial environments where factors such as lighting variations, environmental noise, and cluttered backgrounds significantly impact inspection results. For example, during the production and assembly of vehicle chassis, uneven lighting, reflective objects, and cluttered backgrounds can all interfere with the accurate detection of the vision system, further exacerbating the limitations of traditional methods. Therefore, manual inspection not only struggles to meet the demands of rapid and efficient production but also easily leads to the failure to promptly detect rivet installation defects, affecting production quality.

[0004] Therefore, how to design a vehicle chassis rivet inspection method that can improve the inspection accuracy of vehicle chassis rivets and reduce errors during manual inspection has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for detecting rivets on a vehicle chassis, so as to solve the above-mentioned technical problems in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for inspecting rivets on a vehicle chassis, comprising the following steps: Step S1: Obtain the vehicle chassis image dataset; Step S2: Based on the characteristics of chassis rivet image data, improve the YOLO v9 neural network and construct a backbone network, a multi-path attention mechanism, a residual fusion module, and an output module. Step S3: Input the processed chassis rivet image data into the above neural network for training, and obtain the optimal neural network detection model through backpropagation; Step S4: Use the above neural network detection model to obtain the detection position coordinates of the rivet; Step S5: After comparing the detection position coordinates of the rivet with the set installation position coordinate data, a detection result is generated indicating whether the rivet is installed in place or not.

[0007] Preferably, the specific content of the vehicle chassis image dataset is as follows: Step S11: Deploy multiple camera positions in the vehicle mounting bus to collect vehicle chassis image datasets, classify the datasets according to different environments, and obtain vehicle chassis images containing multiple background types. Step S12: Use image editing tools to preprocess the image containing multiple background types and detect the vehicle chassis area from the multiple background types; Step S13: Use software to mark multiple vehicle chassis areas to obtain a vehicle chassis image dataset: chassis data { }, tag data { };in This represents the nth chassis image. Indicates and The corresponding nth label data, This indicates the total number of data points.

[0008] Preferably, the software used in step S13 is LabelMe.

[0009] Preferably, the backbone network includes a programmable gradient information module and a latent embedding architecture. The programmable gradient information module can optimize feature extraction, and the latent embedding architecture combines the principles of CSPNet and ELAN to optimize information flow through gradient path planning, ensuring efficient inference in multi-scale object detection.

[0010] Preferably, the multipath attention mechanism combines spatial attention and channel attention.

[0011] Preferably, the residual fusion module performs feature fusion by summing the global weights and local weights through short-circuit connections.

[0012] Preferably, the output module employs a non-maximum suppression algorithm to remove redundant detection results.

[0013] Preferably, the processed chassis rivet image data is input into the aforementioned neural network for training, and the optimal neural network detection model is obtained through backpropagation. Step S31: Input the nth chassis image of the above vehicle chassis image dataset into the programmable gradient information of the backbone network of the neural network and the generative latent embedding architecture for object detection to obtain the feature map BackboneRes, whose number of channels becomes 1024. Step S32: Input the feature map BackboneRes into the multi-path network structure and extract the local multi-path feature Local-MultiPathRes, whose channel number becomes 1024 / path, where path is the number of split paths; Step S33: After fusing the local feature Local-MultiPathRes and the global feature MultiPathRes, the result is fed into the residual fusion module for optimization to obtain the feature ResidualRes. After an upsampling process, the number of channels becomes the input size. Step S34: Calculate the loss function L using the backpropagation algorithm and update the network parameters; the formula for calculating the loss function is: ; in: For the sample size, For real labels, For predicting labels; Cross-entropy loss is used for classification. Intersection over Union (IoU) loss is used for bounding box regression; For the true bounding box, To predict the bounding box; As a weighting factor; Step S35: When the number of training iterations reaches a predetermined number T or the loss function L converges to a set threshold, stop training and obtain the optimal neural network detection model.

[0014] Preferably, the specific details of obtaining the rivet detection position coordinates using the above-mentioned neural network detection model are as follows: Step S41: Based on the rivet position coordinates output by the neural network detection model, obtain the corresponding pixel prediction score vector. ,in Indicates the first The prediction score for each pixel, with a value range of . ; Step S42, for Filter the pixels in the data and retain the scores. The pixels, of which For threshold; ; in, This represents the coordinates of the corresponding pixel. The set of high-scoring pixels that are retained; Step S43: Fit the retained rivet position coordinates using geometric methods: If the rivet positions are linearly distributed, then the following formula can be used for linear fitting: ; in, The slope The intercept is calculated using the least squares method; If the rivet positions conform to a circular pattern, then the circle fitting formula is used: ; in, To fit the coordinates of the circle center, The radius is calculated using the Hough transform; Step S44: Use the fitted geometric feature parameters as the detection position coordinates of the rivet.

[0015] Preferably, after comparing the detection position coordinates of the rivet with the set installation position coordinate data, the specific content of the detection result (whether the rivet is installed in place or not) is as follows: Step S51: Set the detection position coordinates of the vehicle rivet as follows: Set the installation location coordinates as Calculate the Euclidean distance between the two coordinates mentioned above. : ; Step S52: Based on the set threshold To determine if the rivets are installed correctly, if... If the rivet is installed correctly, it is considered that the rivet is installed correctly; otherwise, it is considered that the rivet is not installed correctly, and then the test result is output.

[0016] The beneficial effects of this invention are as follows: The vehicle chassis rivet inspection method of the present invention can significantly improve the inspection accuracy of vehicle chassis rivets, avoid the errors of manual inspection in the prior art, and ensure the accuracy and safety of rivet installation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly described below, and the specific embodiments of the present invention will be further described in detail with reference to the drawings, wherein...

[0018] Figure 1 A flowchart of a vehicle chassis rivet detection method provided in an embodiment of the present invention. Detailed Implementation

[0019] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the invention.

[0020] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0021] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0022] In all the examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0023] like Figure 1 As shown, this embodiment of the invention provides a method for detecting rivets on a vehicle chassis, which includes the following steps: Step S1: Obtain the vehicle chassis image dataset; Step S2: Based on the characteristics of chassis rivet image data, improve the YOLO v9 neural network and construct a backbone network, a multi-path attention mechanism, a residual fusion module, and an output module. Step S3: Input the processed chassis rivet image data into the above neural network for training, and obtain the optimal neural network detection model through backpropagation; Step S4: Use the above neural network detection model to obtain the detection position coordinates of the rivet; Step S5: After comparing the detection position coordinates of the rivet with the set installation position coordinate data, a detection result is generated indicating whether the rivet is installed in place or not.

[0024] The vehicle chassis rivet detection method provided in this invention can significantly improve the detection accuracy of vehicle chassis rivets, avoid errors in manual inspection as in the prior art, and ensure the accuracy and safety of rivet installation.

[0025] Furthermore, the specific content of the vehicle chassis image dataset is as follows: Step S11: Deploy multiple camera positions in the vehicle mounting bus to collect vehicle chassis image datasets, classify the datasets according to different environments, and obtain vehicle chassis images containing multiple background types. Step S12: Use image editing tools to preprocess the image containing multiple background types and detect the vehicle chassis area from the multiple background types; Step S13: Use software to mark multiple vehicle chassis areas to obtain a vehicle chassis image dataset: chassis data { }, tag data { };in This represents the nth chassis image. Indicates and The corresponding nth label data, This indicates the total number of data points.

[0026] Preferably, the software used in step S13 is LabelMe.

[0027] Furthermore, the backbone network includes a Programmable Gradient Information Module (PGI) and a Generative Latent Embedding Architecture (GLEAN). The PGI module optimizes feature extraction, while the GLEAN combines the principles of CSPNet and ELAN, optimizing information flow through gradient path planning to ensure efficient inference in multi-scale object detection. The PGI module optimizes the feature extraction process, reducing the accumulation of gradient errors in deep supervision and improving the learning efficiency of multi-branch prediction. GLEAN's lightweight design and flexible layer aggregation mechanism enable the model to enhance its ability to detect objects of different sizes while maintaining inference speed.

[0028] Specifically, the multipath attention mechanism combines spatial attention and channel attention. Spatial attention focuses on image regions, while channel attention weights feature channels, optimizing the model's focus on important features. By using MLP (Multilayer Perceptron), max pooling, and average pooling operations, the most salient regional and global features can be extracted and combined. Dot-multiplication convolution is used to further integrate this information, thereby improving the model's feature capture ability in complex relationships.

[0029] Furthermore, the residual fusion module sums the global and local weights through short-circuit connections to perform feature fusion. This approach avoids the vanishing gradient problem in deep networks, improves information transmission efficiency, and ensures that the fused output includes both global and local information.

[0030] Specifically, the output module uses a nonmaximum suppression algorithm to remove redundant detection results.

[0031] The formula for the output module is: Outputs = NMS(ResidualRes, θ); Wherein, NMS is a non-maximum suppression algorithm, and θ is the threshold of NMS, which is used to control the removal of redundant detections.

[0032] Furthermore, the processed chassis rivet image data is input into the aforementioned neural network for training, and the optimal neural network detection model is obtained through backpropagation. The specific details are as follows: Step S31: Input the nth chassis image of the above vehicle chassis image dataset into the programmable gradient information of the backbone network of the neural network and the generative latent embedding architecture for object detection to obtain the feature map BackboneRes, whose number of channels becomes 1024. Step S32: Input the feature map BackboneRes into the multi-path network structure and extract the local multi-path feature Local-MultiPathRes. Its channel number becomes 1024 / path, where path is the number of split paths, usually set to 16. Step S33: After fusing the local feature Local-MultiPathRes and the global feature MultiPathRes, the result is fed into the residual fusion module for optimization to obtain the feature ResidualRes. After an upsampling process, the number of channels becomes the input size. Step S34: Calculate the loss function L using the backpropagation algorithm and update the network parameters; the formula for calculating the loss function is: ; in: For the sample size, For real labels, For predicting labels; Cross-entropy loss is used for classification. Intersection over Union (IoU) loss is used for bounding box regression; For the true bounding box, To predict the bounding box; As a weighting factor; Step S35: When the number of training iterations reaches a predetermined number T or the loss function L converges to a set threshold, stop training and obtain the optimal neural network detection model.

[0033] The formula for stopping training can be: Stop Training: Iter≥T or L≤ ; Where Stop Training means to stop training, and Iter is the current iteration number. The threshold for the set loss function.

[0034] Furthermore, the specific details of obtaining the rivet detection position coordinates using the aforementioned neural network detection model are as follows: Step S41: Based on the rivet position coordinates output by the neural network detection model, obtain the corresponding pixel prediction score vector. ,in Indicates the first The prediction score for each pixel, with a value range of . ; Step S42, for Filter the pixels in the data and retain the scores. The pixels, of which This is the threshold, usually set to 0.5; ; in, This represents the coordinates of the corresponding pixel. The set of high-scoring pixels that are retained; Step S43: Fit the retained rivet position coordinates using geometric methods: If the rivet positions are linearly distributed, then the following formula can be used for linear fitting: ; in, The slope The intercept is calculated using the least squares method; If the rivet positions conform to a circular pattern, then the circle fitting formula is used: ; in, To fit the coordinates of the circle center, The radius is calculated using the Hough transform; Step S44: Use the fitted geometric feature parameters as the detection position coordinates of the rivet.

[0035] Furthermore, after comparing the detection position coordinates of the rivet with the set installation position coordinate data, the specific content of the detection result indicating whether the rivet is installed in place or not is as follows: Step S51: Set the detection position coordinates of the vehicle rivet as follows: Set the installation location coordinates as Calculate the Euclidean distance between the two coordinates mentioned above. : ; Step S52: Based on the set threshold To determine if the rivets are installed correctly, if... If the rivet is installed correctly, it is considered properly installed; otherwise, it is considered improperly installed, and the test result is output. That is: ; in It is a set threshold that represents the maximum tolerance between the rivet detection position and the set installation position, which usually depends on the accuracy requirements of the specific application.

[0036] This invention acquires a vehicle chassis image dataset; and based on the YOLO v9 neural network architecture and multi-path attention mechanism, uses the backbone network of the YOLO v9 neural network to extract global features and obtain global feature weights; processes the feature map by channel path to extract multi-path local weights; fuses the multi-path local weights with the global feature weights to generate fused weights; inputs the fused weights into a residual network for training to obtain an optimized neural network detection model; inputs the vehicle chassis image data into the detection model to obtain the rivet position coordinates; finally, compares them with the vehicle rivet installation position data to determine whether the rivets are correctly installed.

[0037] This invention utilizes a computer vision-based deep learning model for the automatic detection of rivets on vehicle chassis, significantly improving the detection accuracy and ensuring the accuracy and safety of rivet installation. It also employs a multi-path attention mechanism to fuse global and local features, enhancing detection accuracy, and a residual fusion module to optimize information transmission, thus improving model performance. The introduction of the GLEAN architecture improves the model's real-time performance and adaptability. By combining the multi-path attention mechanism with the residual fusion module, the real-time performance and adaptability of the detection system are effectively improved, meeting the automotive industry's demand for efficient and accurate inspection. Furthermore, this invention optimizes the rivet inspection process through an intelligent detection system, improving production line efficiency and reducing labor costs, thus contributing to the automation and intelligent development of modern automobile manufacturing.

[0038] While specific embodiments of the invention have been described in detail by way of examples, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of the invention. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of the invention. The scope of the invention is defined by the appended claims.

Claims

1. A method for detecting rivets on a vehicle chassis, characterized in that, It includes the following steps: Step S1: Obtain the vehicle chassis image dataset; Step S2: Based on the characteristics of chassis rivet image data, improve the YOLO v9 neural network and construct a backbone network, a multi-path attention mechanism, a residual fusion module, and an output module. Step S3: Input the processed chassis rivet image data into the above neural network for training, and obtain the optimal neural network detection model through backpropagation; Step S4: Use the above neural network detection model to obtain the detection position coordinates of the rivet; Step S5: After comparing the detection position coordinates of the rivet with the set installation position coordinate data, a detection result is generated indicating whether the rivet is installed in place or not.

2. The vehicle chassis rivet inspection method according to claim 1, characterized in that, The specific content of obtaining the vehicle chassis image dataset is as follows: Step S11: Deploy multiple camera positions in the vehicle mounting bus to collect vehicle chassis image datasets, classify the datasets according to different environments, and obtain vehicle chassis images containing multiple background types. Step S12: Use image editing tools to preprocess the image containing multiple background types and detect the vehicle chassis area from the multiple background types; Step S13: Use software to mark multiple vehicle chassis areas to obtain a vehicle chassis image dataset: chassis data { }, tag data { };in This represents the nth chassis image. Indicates and The corresponding nth label data, This indicates the total number of data points.

3. The vehicle chassis rivet inspection method according to claim 2, characterized in that, The software used in step S13 is LabelMe.

4. The vehicle chassis rivet inspection method according to claim 1, characterized in that, The backbone network includes a programmable gradient information module and a latent embedding architecture. The programmable gradient information module can optimize feature extraction, and the latent embedding architecture combines the principles of CSPNet and ELAN to optimize information flow through gradient path planning, ensuring efficient inference in multi-scale object detection.

5. The vehicle chassis rivet inspection method according to claim 1, characterized in that, Multipath attention mechanisms combine spatial attention and channel attention.

6. The vehicle chassis rivet inspection method according to claim 1, characterized in that, The residual fusion module sums the global weights and local weights through short-circuit connections to perform feature fusion.

7. The vehicle chassis rivet inspection method according to claim 1, characterized in that, The output module uses a non-maximum suppression algorithm to remove redundant detection results.

8. The vehicle chassis rivet inspection method according to claim 1, characterized in that, The processed chassis rivet image data is input into the aforementioned neural network for training, and the optimal neural network detection model is obtained through backpropagation. The specific details are as follows: Step S31: Input the nth chassis image of the above vehicle chassis image dataset into the programmable gradient information of the backbone network of the neural network and the generative latent embedding architecture for object detection to obtain the feature map BackboneRes, whose number of channels becomes 1024. Step S32: Input the feature map BackboneRes into the multi-path network structure and extract the local multi-path feature Local-MultiPathRes, whose channel number becomes 1024 / path, where path is the number of split paths; Step S33: After fusing the local feature Local-MultiPathRes and the global feature MultiPathRes, the result is fed into the residual fusion module for optimization to obtain the feature ResidualRes. After an upsampling process, the number of channels becomes the input size. Step S34: Calculate the loss function L using the backpropagation algorithm and update the network parameters; The formula for calculating the loss function is: ； in: For the sample size, For real labels, For predicting labels; Cross-entropy loss is used for classification. Intersection over Union (IoU) loss is used for bounding box regression; For the true bounding box, To predict the bounding box; As a weighting factor; Step S35: When the number of training iterations reaches a predetermined number T or the loss function L converges to a set threshold, stop training and obtain the optimal neural network detection model.

9. The vehicle chassis rivet inspection method according to claim 1, characterized in that, The specific details of obtaining the rivet detection position coordinates using the aforementioned neural network detection model are as follows: Step S41: Based on the rivet position coordinates output by the neural network detection model, obtain the corresponding pixel prediction score vector. ,in Indicates the first The prediction score for each pixel, with a value range of . ; Step S42, for Filter the pixels in the data and retain the scores. The pixels, of which For threshold; ； in, This represents the coordinates of the corresponding pixel. The set of high-scoring pixels that are retained; Step S43: Fit the retained rivet position coordinates using geometric methods: If the rivet positions are linearly distributed, then the following formula can be used for linear fitting: ； in, The slope The intercept is calculated using the least squares method; If the rivet positions conform to a circular pattern, then the circle fitting formula is used: ； in, To fit the coordinates of the circle center, The radius is calculated using the Hough transform; Step S44: Use the fitted geometric feature parameters as the detection position coordinates of the rivet.

10. The vehicle chassis rivet inspection method according to claim 1, characterized in that, After comparing the detected position coordinates of the rivet with the set installation position coordinate data, the specific content of the detection result (whether the rivet is installed in place or not) is as follows: Step S51: Set the detection position coordinates of the vehicle rivet as follows: Set the installation location coordinates as Calculate the Euclidean distance between the two coordinates mentioned above. : ； Step S52: Based on the set threshold To determine if the rivets are installed correctly, if... If the rivet is installed correctly, it is considered that the rivet is installed correctly; otherwise, it is considered that the rivet is not installed correctly, and then the test result is output.

Images