We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
基于改进算法 YOLOv5+的混凝土轨枕裂纹检测.
- Authors
令雅莉; 杨桂芹; 张又元; 王小鹏
- Abstract
Based on the existing research results, an improved algorithm YOLOv5 +is proposed on the basis of insufficient detection efficiency of concrete sleeper cracks. It is mainly based on the YOLOv5 network model to effectively detect concrete sleeper cracks. First, the divide and conquer label strategy is adopted to increase the actual proportion of cracks in the label, so as to solve the problem of large changes in the size of concrete sleeper cracks and make the network more conducive to extracting effective features Secondly, the maximum pool layer of the SPP module in YOLOv5 network structure is changed to the average pool layer to reduce the phenomenon of missing crack detection. At the same time, SE attention module is embedded in YOLOv5 backbone network to improve the detection ability of small cracks. Finally, the new detection scale and feature fusion network are combined to reduce the leakage of micro cracks. The experimental results show that the improved algorithm YOLOv5+based on YOLOv5 network model, except that the recall rate does not change much, improves the accuracy rate Precision by 6. 5%, the average accuracy mean mAP by 8%, and the frame rate FPS is also improved, which can meet the real-time detection requirements.
- Publication
Railway Standard Design, 2024, Vol 68, Issue 4, p70
- ISSN
1004-2954
- Publication type
Article
- DOI
10.13238/j.issn.1004-2954.202210030001