BSF-YOLOv8n: Enhanced Small Object Detection Model

This document is a technical report that presents BSF-YOLOv8n, an optimized model for small object detection in UAV aerial photography, based on the YOLOv8 architecture. The report outlines the challenges faced in UAV target detection, particularly for small objects that often lack distinct features and are easily confused with complex backgrounds. The BSF framework introduces three core enhancements: the Bidirectional Feature Pyramid Network (BiFPN) for improved multi-scale feature fusion, the Squeeze-and-Excitation Network (SENet) for adaptive channel recalibration, and the FASFFHead for resolving cross-scale feature conflicts. Experimental results on the VisDrone2019 Dataset indicate that BSF-YOLOv8n achieves an 8% improvement in mean average precision compared to the baseline YOLOv8n, significantly enhancing detection accuracy for small targets while maintaining real-time performance. The report discusses the implications of these advancements for UAV applications, emphasizing the model's efficiency and effectiveness in complex aerial environments.