Small Object Orientated Detection Model Optimization Based on Lightweight UAV Platform

In light of limited hardware resources and the inherent challenge in detecting small objects, existing object detection models often struggle to strike an optimal balance between accuracy and delay when deployed on UAV platforms. To address this issue, this paper presents a novel approach involving the enhancement and optimization of the YOLOv8n model. According to the need of UAV object detection tasks, we optimize the original detection head to improve reasoning speed while maintaining accuracy. Additionally, a novel efficient feature fusion network structure is designed to enhance the level of detail in the feature map. Through these enhancements, our proposed model aims to address the accuracy-delay trade-off conundrum encountered in object detection on UAV platforms. Experiments on the VisDrone2019 dataset demonstrate that compared with the original benchmark model, this optimized YOLOv8n detection model improves the accuracy with the AP \(_val\) value increasing by 0.2% to 18.2%, while simultaneously increasing the FPS by 17, thereby achieving a good balance between accuracy and delay.