Detection Method for Abnormality in Corridor Environment of Transmission Line Based on Improved YOLOv5

As a pivotal element within the power grid infrastructure, ensuring the safety of the transmission line corridor environment stands as a primary concern. In practical scenarios, an assortment of anomalous objects, each exhibiting unique characteristics, may inadvertently manifest within the transmission line corridor environment, thereby posing significant threats to the stability of the power grid. In this study, we introduce an innovative recognition algorithm based on deep learning, leveraging enhancements to the YOLOv5 framework, specifically designed to detect various objects such as bird nets, cranes, and tower cranes. To bolster recognition accuracy, we integrate attention mechanisms, C2f, and a decoupling head module into the YOLOv5 architecture. Additionally, we introduce a reparameterization module within the decoupling head, effectively reducing the overall parameter count. Through a range of tests conducted on a self-built dataset, we validate the recognition prowess of our proposed deep learning network. Experimental results demonstrate notable enhancements in both accuracy and recall metrics when compared to the baseline YOLOv5 network.