Research on Small Target Detection Algorithm Based on Improved YOLOv5
Xingya Yan, Xiaohuan Li · 2023
Accompanied by nowadays rapid development of artificial intelligence, scientific computing equipment and the upgrading of portable camera equipment, the detection of large and medium targets in videos and images can no longer satisfied the needs of people’s real life, and small target detection has become a fiery-hot spot in the range of target detection research problem. Aiming at the problems of missed detection and low precision caused by factors such as few effective features and low resolution of small targets in the picture, this paper proposes a small object detection algorithm based on improved YOLOv5(You Only Look Once). First of all, in order to solve the problems of missed detection and false detection in small target detection, the SE(Squeeze-Excitation) attention mechanism is integrated into the backbone network to improve the performance of the model. Secondly, in order to solve the unknown geometric transformation problem of the same object in different scenes and different angles, the DCNv2(Deformab1e ConvNets v2) model is added to expand the receptive field and improve the robustness of target detection. The experimental results show that the average detection accuracy (mAP) of the improved YOLOv5s model on the VisDrone dataset has reached 37.56%, which is 5.37% higher than that of YOLOv5s and can effectively detect small targets.