Title: Intrusion detection algorithm for electronic fence in substation area based on YOLOv4

Authors: Bo Liu; Hui Yan; Fanlin Meng; Zhichuang Liu

Addresses: State Grid Jiangsu Electric Power Co., Ltd., Xuzhou Power Supply Company, Xuzhou, 221000, China ' State Grid Jiangsu Electric Power Co., Ltd., Xuzhou Power Supply Company, Xuzhou, 221000, China ' State Grid Jiangsu Electric Power Co., Ltd., Xuzhou Power Supply Company, Xuzhou, 221000, China ' State Grid Jiangsu Electric Power Co., Ltd., Xuzhou Power Supply Company, Xuzhou, 221000, China

Abstract: This study proposes an intrusion detection algorithm for substation electronic fences based on an improved YOLOv4, aiming to enhance real-time monitoring and safety. A pedestrian intrusion detection scheme is developed using YOLOv4, optimised with soft thresholding, depthwise separable convolution, and CBAM to address complex backgrounds, lighting changes, and target variability. The improved model achieved stable performance after 1,300 and 1,200 training iterations on the PFSD and UA-DETRAC datasets, respectively. On the PFSD dataset, it reached a mAP of 97.96%, a recall of 90.63%, an IOU of 94.32%, and 58 FPS. In real-world substation tests, the detection accuracy was 96.74%, demonstrating its effectiveness in enhancing substation security.

Keywords: substation; electronic fence; intrusion detection; YOLOv4; target detection.

DOI: 10.1504/IJPT.2026.154605

International Journal of Powertrains, 2026 Vol.15 No.2, pp.138 - 155

Received: 28 May 2025
Accepted: 18 Oct 2025
Published online: 07 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article