DETECTION OF BIRD'S NEST ON TRANSMISSION LINES FROM AERIAL IMAGES BASED ON DEEP LEARNING MODEL

The bird's nest on transmission lines poses a threat to the safe operation of transmission equipment and even affects the stability of the entire power system. Recent-ly, with the rapid development of 5G technology, unmanned aerial vehicle (UAV) tech-nology, and artificial intelligence technology, intelligent patrol transmission lines based on UAVs have become an inevitable trend in the development of power inspection. To ad-dress the problems of low recognition accuracy and low Recall of bird's nest detection in complex backgrounds by traditional methods, an improved YOLOv3 automatic detection model of bird's nest based on attentional feature fusion (AFF-YOLOv3) is proposed in this paper. The model first adds an attentional feature fusion network to the YOLOv3 top-down sampling process, calculates semantic weights based on the deep-level feature map, and then uses the semantic weights as a guide for selecting low-level features to ob-tain more valuable low-level features. Finally, the selected low-level feature maps and the high-level feature maps are concatenated to obtain robust features with both location in-formation and semantic information. The experimental results show that AFF-YOLOv3 achieves 87.58% average precision (AP) on the transmission line bird's nest dataset, and the model has stronger generalization ability and applicability compared with other detec-tors.