Research on Remote Sensing Small Object Detection Based on Improved YOLOv8

This study introduces SSE-YOLO, a lightweight extension algorithm derived from the YOLOv8s framework. It addresses two issues: model weight reduction and real-time performance for small target detection in UAV remote sensing. By reconstructing the backbone network with ShuffleNetv2 and optimizing the neck network via a Slim-neck module based on GSConv and VoVGSCSP, the algorithm achieves significant reductions in both parameters and computational demands. Further improvements include the integration of the ECA module to boost feature representation and the adoption of the Wise-IoU loss function for precise localization. Experimental evaluations reveal that SSE-YOLO reduces model's size and computational load by 47.7% and 53.6%, respectively, compared to YOLOv8$s$, while maintaining high accuracy with only minor declines in mAP@0.5 (3.6%) and mAP@0.5:0.95 (8.8%). The results demonstrate its effectiveness as an efficient, lightweight solution for UAV-based detection of small targets.