Transmission Tower Detection Scheme Based on STPP-YOLO Network

Automated inspection of transmission line infrastructure using robots or drones is an important way to ensure transmission safety. The automatic detection of transmission tower is the premise of power grid inspection automation. In this paper, a transmission tower detection scheme using a lightweight Swin Transformer-based PP-YOLO network (STTP-YOLO) is proposed. It can locate the tower target from aerial images while balancing the detection speed and accuracy. First, a dataset of transmission towers suitable for deep learning research is collected, and accurate labeling is completed manually to improve the study of the model. Second, a new tower backbone network is proposed to replace the ResNet50vd-DCN in the original model by combining the swin transformer with the traditional convolutional neural network, and the model can extract the tower information more accurately. Then, the Feature Pyramid Network (FPN) is used for decoding layer construction to perform simple and fast multi-scale feature fusion to extract target information more accurately before result prediction. Finally, the experimental results prove that the proposed scheme has strong competitiveness in accuracy and detection speed, which can be deployed in the task of automatic detection pole tower of power grid, and has a good guiding effect on the detection of auxiliary components of transmission tower.