Road Target Detection Method Based on Improved Lightweight YOLOv5s
Cen Gao, Die Hu · 2024
Aiming at the current YOLOv5s on the edge devices are susceptible to energy consumption and memory, which makes the traditional YOLO series algorithms recognition efficiency is low, for this reason, this paper proposes a road target detection algorithm based on the improved YOLOv5s. The algorithm adopts the WIoU loss function with a two-layer attention mechanism to raise the network's localisation accuracy of the target; adds the SE attention mechanism to the tail of the backbone network to help the network obtain important information; uses the GhostNet network to replace some of the backbone network convolutional layers in the original YOLOv5s network making the model less parametric. The experimental results show that the improved YOLOv5s has a 2.3% improvement in degree of preciseness in detection and the number of parameters has decreased by about 7% tested on the KITTI dataset, which makes it more practical.