Method for detecting position and type of workpiece on subway vehicle side inspection image
A subway and workpiece technology, applied in the field of machine vision industrial inspection, can solve a large number of manual operations and other problems, achieve the effect of reducing labor intensity, reducing difficulty, and facilitating the analysis of workpiece status
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Example Embodiment
[0029]Example one:
[0030]The present invention is a method for detecting the position and type of workpieces on the side inspection image of a subway car. The model training embodiment mainly describes that in order to obtain a workpiece detection model, the image data acquisition of the subway car body, the target labeling of the workpiece in the image, and the construction of the YOLOv4 deep learning network are mainly described. And training, and model output, the specific implementation process is as follows:
[0031]1) Install two industrial cameras on both sides of the subway track to scan the body of the subway train and output the original scanned picture with a resolution of 65535×1808, and divide the original picture into sub-pictures of 1808×1808 size to save.
[0032]2) Use the Colabeler labeling tool to label each workpiece appearing on the picture. In this embodiment, it mainly includes different screws, upper brake pads, lower brake pads, and iron wires. There are a total of...
Example Embodiment
[0035]Embodiment two:
[0036]Such asfigure 1 As shown, the invention constructs a subway train workpiece detection model based on the YOLOv4 target detection network architecture. YOLO is a target detection model based on deep learning. The present invention builds a new subway workpiece detection model based on YOLOv4. The detection algorithm flow is as followsfigure 1 As shown, the steps of model training are as follows:
[0037]a) First use the pre-trained weights of the CSPDarknet53 network to initialize the YOLOv4 backbone network. The weights are trained using MS COCO target detection data, which can detect 80 objects in the MS COCO dataset. In order to adapt to the application of subway body workpiece detection, Modify the output category of the YOLOv4 network to the number of types of artifacts (10 in the present invention), so as to meet the conditions of migration learning;
[0038]b) For a sub-picture to be trained, its resolution is 1808×1808, and the resolution of the picture n...
Example Embodiment
[0051]Embodiment three:
[0052]After the above-mentioned training method is trained on the subway car body workpiece detection data set, the present invention can obtain a trained workpiece detection model, which can run under the Darknet deep learning framework and output a length of S×S×(B×5+ The prediction tensor of C), in order to detect the position of each workpiece and identify the type of the workpiece, the present invention introduces a non-maximum suppression algorithm (NMS) to find the target frame and confidence level from the prediction tensor. The steps are as follows:
[0053]1) Set a threshold to filter out all bounding boxes of target candidate frames smaller than this threshold;
[0054]2) Select the one with the highest target confidence of the bounding box of a certain workpiece candidate frame, which is box_best;
[0055]3) Calculate the IOU (Intersection over Union) between box_best and other candidate frame bounding boxes of other workpieces, that is, the intersection of...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap