Low-Power Portable System for Power Grid Foreign Object Detection Based on the Lightweight Model of Improved YOLOv7
Yonghuan He, Rong Han Wu, Chao Dang · IEEE Access · 2024
Foreign objects caught and entangled on power grid transmission lines, power towers and other equipment can pose potential threats to the power system. The detection algorithm for foreign objects in the power grid cannot achieve the optimal balance in terms of accuracy and efficiency. The increasing demand for embedded devices in practical applications has led to the development of efficient, lightweight networks as a trend. In order to meet the actual power grid foreign object detection needs, this paper proposes a low-power portable power grid detection system based on lightweight improved YOLOv7, which is designed to be deployed on low-power portable embedded devices while ensuring a balance between detection efficiency and accuracy. Through the parallel module of segmentation and detection, the model completes instance segmentation while detecting, reducing the missed detection of object bounding boxes and increasing the attention to small objects. In addition, lightweight improvements are made to each branch of the network to ensure detection efficiency while significantly reducing the amount of calculation. Furthermore, a professional power grid foreign object detection data set is proposed, and the proposed model is experimentally verified on the dataset. The experimental results demonstrate that the model proposed has higher detection performance in foreign object detection in power grids. Finally, the model is deployed on the Jetson Nano embedded device and showed superior adaptability and inference speed, proving the practicability of the entire system for foreign object detection in the power grid.