The content of this article comes from the "Surveying and Mapping Bulletin" No. 8, 2024, drawing review number: GS Jing (2024) No. 1527
Optimization and application of deep learning model for metro tunnel disease detection
You Xiangjun1, Zhao Xia2, Long Sichun3, Wang Jiawei1, Zheng Ying2, Kuang Lijun4
1. Zhejiang Huazhan Research & Design Institute Co., Ltd., Ningbo 315000, Zhejiang, China; 2. School of Computer and Artificial Intelligence, Beijing Technology and Business University, Beijing 102446, China; 3. School of Earth Sciences and Spatial Information Engineering, Hunan University of Science and Technology, Xiangtan 411201, Hunan, China; 4. China Construction Fifth Engineering Bureau Co., Ltd., Shenzhen, Guangdong, 518108
Funds: National Natural Science Foundation of China (42377453; 41877283); Hunan Science and Technology Innovation Program (2021RC4037; 2033JJ30235); Research Project of Hunan Provincial Department of Natural Resources (2021-18); 2024 Ningbo "Science and Technology Yongjiang 2035" Key Technology Breakthrough Program Scientific Research Project
Keywords: deep learning, model optimization, detection methods, tunnel disease
Citation format: YOU Xiangjun, ZHAO Xia, LONG Sichun, et al. Optimization and Application of Deep Learning Model for Disease Detection in Metro Tunnels[J]. Bulletin of Surveying and Mapping, 2024(8): 96-101.doi: 10.13474/j.cnki.11-2246.2024.0817
Abstract:In this paper, aiming at four common diseases in subway tunnels, including water leakage, cracks, structural plastering cracking and spalling blocks, the disease detection method of subway tunnels based on lidar scanning point cloud data and deep learning was studied. Firstly, the ACmix attention module was introduced to make the network take into account both global and local features, and improve the detection effect of small targets such as cracks and cracks. Then, the regression loss function was optimized to improve the convergence smoothness and regression accuracy, and reduce the detection error. Finally, the integrated generation of orthophoto image preprocessing, batch detection, result fusion and detection result report was realized, so as to improve the disease detection rate of large-scale orthophoto projection map. The experimental results show that under the condition of selecting the IoU threshold of 0.5, the accuracy of the improved YOLOv8 algorithm in the tunnel disease test and detection is increased from 90.65% to 91.18%, which basically realizes the intelligent detection of four common diseases in subway tunnels based on LiDAR scanning, and has been successfully applied in the actual tunnel operation and maintenance engineering.
About authorAbout author:YOU Xiangjun (1978—), male, master, senior engineer, main research direction is engineering surveying and intelligent surveying and mapping. E-mail:[email protected] Corresponding author: Zhao Xia. E-mail:[email protected]
First instance: Yang Ruifang review: Song Qifan
Final Judge: Jin Jun
information