Transmission Line Detection Method Based on Improved Res2Net-YOLACT Model

In order to enhance the efficiency and safety of unmanned aerial vehicle (UAV) inspections in power grids, reduce labor costs, and propose an improved detection method for transmission lines, an enhanced Res2Net-YOLACT method is introduced. Firstly, the feature extraction capability of the YOLACT model based on the Res2Net backbone is strengthened. Next, the task is divided into two parallel tasks using the Feature Pyramid Network (FPN) to predict each mask coefficient vector. Finally, the corresponding detection results are generated through Non-Maximum Suppression (NMS) with DIoU (Distance-Intersection over Union). Experimental results conducted on real public datasets demonstrate that the proposed model improvement can enhance the accuracy of transmission line detection while ensuring the detection speed of the model.