INTELLIGENT DETECTION METHOD FOR FOREIGN OBJECTS ON OVERHEAD LINES OF URBAN BUILDINGS
Keywords:
Urban overhead lines, Fault detection, Improved EfficientDet, Improved Vision Transformer (ViT)Abstract
To address practical engineering challenges of overhead power lines near urban residential buildings, including low erection height, close proximity to building facades, lightweight litter such as plastic bags easily tangled on cables, complex background interference from building walls, vegetation and outdoor air-conditioning units, as well as variable shapes of tangled foreign objects, tiny target sizes and imbalanced scarce fault samples, this paper proposes an intelligent foreign object detection method for overhead lines combining improved EfficientDet and improved Vision Transformer (ViT). Firstly, the improved EfficientDet is adopted to locate regions containing bird nests and tangled plastic bags. Dilated convolutions are embedded into shallow backbone layers P1 and P2 to enlarge the receptive field, and the CBAM attention mechanism replaces the original SE module to strengthen the extraction of contour features of foreign objects under cluttered backgrounds. Secondly, the improved Vision Transformer performs refined discrimination on the presence or absence of foreign objects based on cropped local images. The rigid hard patch embedding of raw images is substituted by multi-layer small convolutions, the number of multi-head attention heads is optimized, and Focal Loss is introduced to alleviate sample imbalance. Actual test results demonstrate that the mean average precision (mAP) of foreign object detection reaches 96.41%, and the classification accuracy of foreign objects is 96.73%. The proposed method exhibits outstanding stability in typical urban scenarios including backlight, building occlusion and dense shrub interference.References
[1] Zang Peng. Research on Fault Location Method of Cable-Overhead Line Hybrid Transmission Line. Xi'an University of Science and Technology, 2022.
[2] Hu Jiajun. Research and Implementation of Target Detection System Based on Bidirectional Feature Fusion EfficientDet. University of Electronic Science and Technology of China, 2023.
[3] Liang Yubin, Shen Rongfeng, Cheng Chengmao, et al. Weed Detection Model of Young and Middle-Aged Mountain Forest Based on Improved YOLOv8. Forest Engineering: 1-12 [2026-06-23].
[4] Aulton C, Chiu Y C, Chiou Y S. Advancing functional reach assessments: a comparison of YOLOv8 human pose estimation and 3D motion capture in a young healthy cohort. Journal of Biomechanics Open, 2026, 1(2): 100007.
[5] Mela L J, Sánchez G C. Integrating dynamic convolution and channel attention into YOLOv8 for enhanced maritime vision. Digital Signal Processing, 2026, 182: 106258.
[6] Talu H M, Baybars C S, Aboalqaraya R, et al. Automated Detection of Taurodontism in Panoramic Radiographs Using a YOLOv8-Based Deep Learning Model. Journal of Imaging Informatics in Medicine, 2026: 1-14 (Prepublish).
[7] Zeng Zhi, Li Xiaofeng, Wu Jialu, et al. Foreign Object Detection Method for Transmission Lines Based on Improved YOLOv8. Software Guide: 1-16[2026-06-23].
[8] Assad S, Isa M A N, Saleh M A S. Hybrid CNN-Transformer models for industrial defect detection: A systematic review. Results in Engineering, 2026, 29: 109457.
[9] Xie Guobo, Xia Wei, Lin Zhiyi, et al. Research on Transmission Line Foreign Object Detection Based on AAGV-YOLOX Model. Journal of Guangdong University of Technology, 2026, 43(2): 81-90.
[10] Xue Ang, Jiang Enyu, Zhang Wentao, et al. Foreign Object Detection in Transmission Line Corridors Based on Fusion of Window Self-Attention Network and YOLOv5. Journal of Shanghai Jiao Tong University, 2025, 59(3): 413-423.
[11] Song Liye, Liu Shuai, Wang Kai. Power Grid Component and Defect Recognition Method Based on Improved EfficientDet. Transactions of China Electrotechnical Society, 2022, 37(9): 2241-2251.