MW-YOLO: Improved YOLOv8n for Lightweight Dense Vehicle Object Detection Algorithm

In recent years, the transportation field has focused on advanced object detection algorithms to develop autonomous vehicles. However, the problem of misidentification and missed detection often occurs due to the high overlap of small size targets on the road. To solve this problem, this paper pro- poses a lightweight dense vehicle object detection algorithm MW-YOLO(Multi-dimensional feature fusion and WIoU loss function YOLO) based on YOLOv8n model. Firstly, a small target detection layer is introduced in the neck network, along with the addition of a detection head, to enhance the network’s feature extraction and detection capabilities for small targets. Subsequently, two multi-scale feature fusion modules, MSFB and HFF, are proposed to enhance the fusion effect of different scale features and reduce the input and output channels. Finally, the WIOU loss function is used to replace the original CIOU function used in YOLOv8n to address the issue of high overlap of road targets. Experimental results on the RAIR-V2X autonomous driving dataset demonstrate that the proposed MW-YOLO model outperforms the original model, with improvements of 5.9% in AP50 and 6% in AP, and the model memory is only 5.1MB, meeting the requirements for deploying models on edge devices.