Research on lightweight small target detection algorithm from the perspective of UAV

Aiming at the problem that the existing small target detection algorithms in the UAV vision system cannot take into account the detection accuracy and real-time detection at the same time, this paper proposes a lightweight UAV vision small target detection algorithm TDOD-YOLO based on YOLOV5s, which firstly takes the YOLOv5s feature extraction layer and the output detection layer as the backbone network and the head network, and then introduces MobileNetv3 to reconfigure the original backbone network. Lightweight network to reconfigure the original backbone network to reduce the model size; secondly, the attention mechanism of Bneck in the backbone network is modified to CBAM, so that the network model pays more attention to the target features, and finally, the loss function is modified to Focal-EIOU to accelerate the convergence speed of the model and further improve the model accuracy. The experimental results show that the average detection accuracy of the TDOD-YOLO algorithm proposed in this paper reaches 33.8%, and compared with YOLOv5s, the average accuracy mAP of the network improves by 1.3% while the amount of parameters is reduced by 42.8% and the amount of computation is reduced by 39%, which proves that the algorithm reduces the size of the model dramatically, and improves the speed of detection while maintaining a good detection performance.