YOLO-Based Multi-Scale Lightweight Insulator String Defect Detection Network

EasyChair Preprint no. 11404

Abstract

When the defect detection model of transmission line insulator strings is deployed on edge devices such as drones, it is essential to condense the model as much as possible while increasing computation speed while maintaining accuracy. This paper proposes a new lightweight target detection algorithm based on YOLOv5, improves the C3 network structure by introducing multi-scale feature information interaction and reducing redundant channel information, and generates two modules that can consider both speed and accuracy, namely FasterC3 and Res2C3. Experiments have shown that combining these two modules can cut the number of model parameters and operations per second by 12%. Furthermore, the computation performance is faster than specific standard lightweight networks with fewer layers.

Keyphrases: lightweight, object detection, power systems, Smart Grids, YOLOv5

@Booklet{EasyChair:11404,
  author = {Xiaosong He and Xiao Wu and Jun Peng and Yuanmin He and Haojun Dai and Xinkai Ma},
  title = {YOLO-Based Multi-Scale Lightweight Insulator String Defect Detection Network},
  howpublished = {EasyChair Preprint no. 11404},

  year = {EasyChair, 2023}}