YOLOv8-GDC: A Lightweight Attention-based Detector for Small Object Detection in UAV-based Emergency Rescue

Object detection using Unmanned Aerial Vehicles (UAVs) is critical for time-sensitive applications such as emergency search and rescue. However, the limited computational resources onboard UAVs pose a significant challenge to deploying complex models, while the accurate detection of small, occluded, or low-resolution targets remains a persistent problem. To address this trade-off between performance and efficiency, this paper proposes YOLOv8-GDC, a lightweight and effective object detector built upon the YOLOv8n baseline. Our core contribution is the novel Gaussian-Depthwise Convolution (GDC) module, a lightweight, implicit attention mechanism designed to enhance the feature representation capabilities for challenging targets. The GDC module strategically combines standard and depthwise convolutions with the GELU activation function to dynamically re-weight features, thereby improving sensitivity to small objects with zero additional parameters or computational overhead. We constructed a specialized dataset for emergency rescue scenarios, encompassing diverse disaster environments and challenging conditions, to validate our approach. Experimental results demonstrate that YOLOv8-GDC achieves a mean Average Precision (mAP50) of 53.2%, outperforming the baseline YOLOv8n by 4.0 percentage points and significantly improving recall by 4.0 points. Crucially, these accuracy gains are achieved with no increase in model parameters, GFLOPs, or significant inference latency, making YOLOv8-GDC a highly practical solution for deployment on resource-constrained UAV platforms in critical missions.