A Structural Information Aided Method for Intelligent Detection of Power Line Targets

Huang Zhencang,Jiang Renjie,Li Dong

DOI: https://doi.org/10.1109/itnec56291.2023.10082674

2023-01-01

Abstract:Accurate real-time detection of low-altitude threat elements is the basis and prerequisite for the safe flight of low-altitude helicopters and UAVs; however, typical low-altitude threat elements such as power lines are low-observable targets, rendering the detection performance of existing methods limited. In this paper, a new structural information-aided power line depth network detection model is proposed by exploiting the spatial location structure information among the power line elements. Firstly, in order to exploit the spatial structure information among threat elements such as power lines, a regression localization network module is designed to obtain the spatial location information of threat elements such as power lines based on the YOLOv3 (You Only Look Once version3) model; then, a relational suggestion network is used to fuse the spatial multi-scale features of power lines and structural feature information to achieve accurate detection of low-observable power line targets. Compared with existing methods, the proposed method improves the power line detection accuracy by about 5% due to the simultaneous mining and exploitation of structural information and spatial multi-scale features of power lines; finally, the effectiveness and advantages of the proposed method are verified by the constructed sample database.