Cavity as a typical road hidden disease is the main cause of roadbed collapse, the development of intelligent transportation for rapid, efficient and accurate identification of cavity disease put forward high requirements. Commonly used ground-penetrating radar nondestructive testing is affected by the road structural layer of complex media and disease morphology of variable coupling, there are issues such as poor consistency of the disease feature extraction and difficult to recognize. In this paper, we took the 100-kilometer measured image data of the highway as a real sample, and proposed an intelligent detection algorithm of road cavity disease based on DFES-YOLOv5s. To address the difficult problem of extracting features of disease with variable morphology, a deformable convolutional alternative C3 module was designed based on the YOLOv5s model to enhance the adaptability to the variable features, and the key features of the disease were effectively extracted; To address the problem of high similarity between disease features and background noise, the Focal-EIOU loss function was used to focus on high-quality anchor frames and comprehensively evaluate the width-height difference, which solved the problems of positive and negative sample imbalance and difficult model convergence; For the problem of many background interference information, the SimAM attention mechanism was added to filter the interference information to improve the performance of the model in the target feature representation and discrimination ability. Experimental results show that the accuracy, average precision, and recall of the DFES-YOLOv5s detection algorithm designed in this study reach 84.2%, 88.5%, and 84.2%, respectively, which are improved by 9.7, 5.3, and 7.0 percentage points, respectively, compared with the conventional YOLOv5s algorithm. This research provides key technical support for enhancing the intelligent detection level and engineering popularization and application of road cavity disease.
ZHAOK, HANQ, ZHANGC B, et al. Deep hough transform for semantic line detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(9): 4793-4806.
XUQiwen, TANGZhenmin, YAOYazhou. Research on image mosaic based on improved SURF algorithm [J]. Journal of Nanjing University of Science and Technology, 2021, 45(2): 171-178.
[7]
KANGM S, KIMN, IMS B, et al. 3D GPR image-based UcNet for enhancing underground cavity detectability[J]. Remote Sensing, 2019, 11(21): 2545.
JIANGLulu, YINYi, MENGZihan, et al. Automatic recognition technology of underground cavities and pipelines in ground penetrating radar images based on YOLO v5 [J]. Traffic Engineering and Technology for National Defence, 2024, 22(5): 7-11. (in Chinese)
[10]
LIY H, ZHAOZ X, LUOY F, et al. Real-time pattern-recognition of GPR images with YOLO v3 implemented by tensorflow[J]. Sensors(Basel), 2020, 20(22): 6476.
[11]
QIUZ, ZHAOZ X, CHENS J, et al. Application of an improved YOLOv5 algorithm in real-time detection of foreign objects by ground penetrating radar[J]. Remote Sensing, 2022, 14(8): 1895.
[12]
LUOT X H, ZHOUY F, ZHENGQ Z, et al. Lightweight deep learning model for identifying tunnel lining defects based on GPR data[J]. Automation in Construction, 2024, 165: 105506.
ZHOUZhong, ZHOUShirong, LIShishuai, et al. Intelligent identification of inner defects in tunnel lining based on improved YOLOv7 [J/OL]. Journal of the China Railway Society(2024- 07-23)[2024-11-20]. in Chinese)
[15]
OMWENGAM M, WUD L, LIANGY, et al. Cognitive GPR for subsurface object detection based on deep reinforcement learning[J]. IEEE Internet of Things Journal, 2021, 8(14): 11594-11606.
[16]
HUH B, FANGH Y, WANGN N, et al. Defects identification and location of underground space for ground penetrating radar based on deep learning[J]. Tunnelling and Underground Space Technology Incorporating Trenchless Technology Research, 2023, 140: 105278.
[17]
REDMONJ, DIVVALAS, GIRSHICKR, et al. You only look once: Unified, real-time object detection [DB/OL]. (2016-5-19)[2024-11-20].