Improved Vehicle and Pedestrian Detection from UAV Perspective Based on Lightweight YOLOv8n

Facing the challenges of vehicle detection from a UAV perspective, such as small targets, occlusions, and dense arrangements, this paper proposes a lightweight vehicle and pedestrian detection algorithm based on YOLOv8n: FDT-YOLOv8. In the neck, the original PA-FPN structure is replaced with the Focusing Diffusion Pyramid Network (FDPN) to enrich the features of contextual information. Secondly, the Task Align Dynamic Detection Head (TADD-Head) is used to replace the original Decoupled-Head, which has been proven to enhance the detection head’s localization and classification performance. By employing shared convolutions, the number of parameters can be significantly reduced, making the model more lightweight. Experimental results show that the model’s parameter count is reduced by 13%, and [email protected] is increased by 4%, achieving the requirement of lightweight design.