A Lightweight Small Object Detection Method for Birds Nest on Electric Tower Based on Attention Mechanism

Abstract

Foreign objects such as bird nests pose significant risks to the safety of transmission lines. Effective inspection and removal of these nests using aerial photography are essential. However, current target detection models face challenges in deployment on embedded devices and in small target detection. This study introduces a lightweight foreign object detection method for transmission lines based on the LW-YOLOv7 algorithm. Enhancements include a small target detection layer in the neck region and a lightweight CBAM module to improve feature extraction. The Ghost module replaces the ELAN and RepConv modules, reducing the model’s weight and easing deployment. Additionally, the WIoU loss function is employed to enhance detection accuracy. Using drone-collected aerial images for training and testing, the proposed method achieves an AP of 89.31% and an accuracy of 96.28% in detecting bird nests. The method outperforms advanced target detection models in terms of model size and detection speed, providing a practical solution for ensuring the safe operation of power systems.