FLF_YOLO: an Efficient and Lightweight Network for Small Target Detection in Aerial Photography

To address the issues related to the ineffective identification of small targets in aerial images and the absence of real-time detection, we propose a lightweight algorithm FLF_YOLO based on the improvement of YOLOv5s.Firstly, FasterNet, an efficient lightweight network, is used instead of the original backbone network, which reduces the computational complexity and improves the inference speed; then, a bottleneck structure combining with LSKNet is designed, which dynamically adjusts the spatial receptive field to improve the detection accuracy in limited resources; finally, Proposing the
$\mathbf{Focal}- \mathrm{a}$
EIoU1.5 loss for adaptive enhancement of bbox regression accuracy through Power transform. Experiments on the open dataset VisDrone 2019 show that in comparison with the baseline model YOLOv5s, the inference speed is improved by 3%, mAP@0.5 is unchanged, and mAP@0.5:0.95 is increased by 1% under the premise of reducing the number of parameters and calculation amount by 37.3% and 37.7% respectively.