A Dual-Platform Human Detection Framework Using Remote Sensing and Deep Learning

Human detection in rural environments poses significant challenges due to variable terrain, limited visibility, and the need for real-time responsiveness. This study investigates the performance of two prominent object detection algorithms, YOLOv8 and Faster R-CNN, applied to high-resolution UAV imagery and medium-resolution satellite data for the purpose of identifying individuals in remote areas. A custom dataset comprising diverse rural scenarios—such as open fields, forested regions, and occluded terrains—was constructed and used to train and evaluate both models. Performance was assessed using standard object detection metrics, including precision, recall, F1score, and mean Average Precision (mAP), as well as computational metrics such as inference time and frames per second (FPS). The results demonstrate that YOLOv8 outperforms Faster R-CNN across all key metrics, achieving a precision of 0.932 and mAP@ 0.5 of 0.947 on UAV imagery while maintaining significantly faster inference speeds suitable for real-time deployment. While both models exhibited reduced performance on satellite images due to resolution constraints, YOLOv8 maintained higher robustness and lower false detection rates. Overall, the findings support the use of YOLOv8 for UAV-based human detection systems in rural areas, offering an effective balance between accuracy and efficiency for both search-andrescue and security applications. Future work will explore multimodal data integration and further optimization for embedded platforms.