MFYOLO: An Improved YOLOv5 Algorithm for Small Target Detection in UAV-Captured Images

With the advancement of unmanned aerial vehicles, the detection of small targets has become a crucial task. However, it is challenging due to the complexity of the scenes and the small size of the targets. In recent years, various object detection methods based on convolutional neural networks (CNN) have made significant progress in small target detection. Despite these advancements, these methods still suffer from a high false detection rate, as well as substantial time and computational overhead, resulting in unsatisfactory performance. Therefore, we propose a multi-fusion YOLOv5 method (MFYOLO) in this paper, which incorporates attention blocks and a multi-scale feature fusion structure to enhance feature fusion and extraction capabilities. To evaluate our proposed method, we conducted extensive comparative experiments on the challenging VisDrone2019 dataset. The experimental results demonstrate that the mAP@0.5 is significantly improved by 10.0% from the original.