Boosting Detection Accuracy: An Enhanced YOLOv8 for Small Target Detection in Remote Sensing

Object detection in remote sensing images has broad applications in military reconnaissance, urban planning, and disaster management. However, detecting small targets and extracting multi-scale features in complex scenes remain challenging. This paper presents YOLOv8s-Improved, enhancing small-target detection via PPHGNetV2, Progressive Feature Pyramid Network (AFPN-P2), and Diverse Branch Block (DBB) modules. Experiments on the DIOR and VisDrone2019 datasets show that YOLOv8s-Improved achieves mAP scores of 0.824 and 45.3%, respectively, representing improvements of 1.5 and 6.4 percentage points over the baseline YOLOv8s model (0.809 and 38.9%). The improved model demonstrates strong performance in multi-category object detection, particularly in complex scenes. The results suggest that the proposed method addresses the challenges of small target detection in remote sensing and exhibits generalization capabilities across different datasets.