YOLOv8n-ESK: Remote Sensing Image Object Detection Algorithm Integrating the ESK Module and Attention Mechanism

This paper presents a YOLOv8n based detector for remote sensing object detection under large scale variation, orientation diversity, and cluttered backgrounds. The method integrates two lightweight modules. First, a scale aware axial separable kernel module named ESK replaces standard convolutions in C2f to strengthen multi scale representation at lower computational cost. Second, a scale adaptive multi path coordinate attention with residual and temperature gated design named SAMPCA_RTD is inserted in the backbone to emphasize direction aware and content relevant features while avoiding over suppression. Experiments on DIOR, RSOD, and PASCAL VOC show consistent improvements. On DIOR the model attains mean average precision with an iou threshold of 0.5(mAP@0.5) of 77.8 compared with 76.6 for YOLOv8n and mAP@0.5-0.95 of 55.0 compared with 53.5 for YOLOv8n with a reduced compute footprint. Ablation studies indicate complementary effects in which SAMPCA_RTD improves attention to small and cluttered targets and ESK enhances scale adaptivity under tight compute budgets.