Lightweight object detection algorithm of UAV aerial image based on improved YOLOv8

In Unmanned Aerial Vehicle image detection, challenges arise from small target sizes, a wide variety of categories, and limited feature information. These factors lead to several issues with existing detection algorithms for small targets, including missed detections, false positives, an excessive number of model parameters, and high computational complexity. Consequently, these algorithms struggle to meet the demands for end-to-end real-time detection in industrial applications and other complex scenarios. Based on the YOLOv8n object detection framework, we propose a lightweight UAV-YOLOv8n detection algorithm for Unmanned Aerial Vehicle images. Firstly, in the neck part, the lightweight BiFPN fusion module is introduced to enhance the multi-scale feature fusion capability of the model and improve the detection accuracy of small targets. Finally, Inner-SIoU is used to optimize the original network loss function, a ratio mechanism is introduced to accelerate the convergence of the model and improve the generalization ability of the model. The prediction accuracy reached 46.7% on the VisDrone2021-DET dataset of public UAV aerial photography. Compared with the YOLOv8n algorithm, accuracy improved by 2.5%, and the number of parameters decreased by 34%. The improved algorithm effectively balances model lightweight, performance and raises a new idea for better deployment in terminal equipment.