Detection of Flying Birds in Airport Monitoring Based on Improved YOLOv5

Abstract

Flying birds affect the safety of aircraft, and it’s difficult to effectively detect and discriminate bird targets because of their small sizes in large-field monitoring. To solve the problem of insufficient feature information of tiny targets and improve the detection performance, in this paper we introduce a method of channel attention mechanisms into the YOLOv5. By modeling the interdependence between channels, the proposed method adaptively learns the weights, to calibrate the feature responses between channels, guides the model to pay more attention to the features with abundant information, and finally improves the accuracy of tiny target detection. We also setup a measured dataset of tiny birds by taking images with optical equipment deployed in airports. The experimental results show that the improved model achieves a certain improvement in detection accuracy and recall rate compared with the original YOLOv5 algorithm.