Hidden Danger Detection Around Power Transmission Lines Based on Improved YOLOv5

Timely detection of hidden dangers around power transmission lines is particularly important in the power field. With various challenges such as few data, large image sizes, complex backgrounds, and small object scales, accurate detection of hidden dangers around power transmission lines remains difficult. In this paper, we propose an effective approach to hidden danger detection around power transmission lines by improving YOLOv5. Firstly, by converting global detection to local detection, an Equal Size Slicing Module (ESSM) is presented to increase the detection ability of small objects and the generalization ability of the model. Secondly, an Adaptive Small Object Enhancement Module (ASOEM) that integrates feature elimination, intact information fusion, attention mechanism, and adaptive feature fusion is used to improve the localization and recognition performance of small objects. We build a dataset of hidden dangers around power transmission lines and verify the validity of our approach on this dataset. The experimental results show that our approach outperforms several state-of-the-art ones. In particular, compared with YOLOv5, the mAP of proposed approach is improved by 8.4%.