Spatial-channel reconstruction for efficient multiscale attention in robotic object detection

Abstract

Real-time object detection is a core capability for autonomous robots, unmanned aerial vehicles (UAVs), and self-driving systems operating in resource-constrained environments. This paper presents spatial-channel enhanced multiscale attention (SCEMA), a novel lightweight attention module designed to enhance robotic perception while minimizing computational overhead for embedded deployment. SCEMA employs a parallel dual-branch architecture that synergistically combines spatial-channel reconstruction with multiscale attention mechanisms. When integrated into the YOLOv8n framework (3.01M baseline parameters), the proposed YOLO-SCEMA model achieves significant performance gains across multiple challenging benchmarks relevant to robotics automation. Experiments on an NVIDIA RTX 4080 GPU demonstrate that on the ExDark dataset, YOLO-SCEMA improves mAP@50 by 7.37% over the baseline (69.07% to 76.44%) while reducing parameters by 36.88% (3.01M to 1.90M) and computational cost by 8.64% (8.1 to 7.4 GFLOPs). Consistent improvements are also observed on VisDrone2019 (+3.24% mAP@50) and FYP (+1.50% mAP@50) datasets. Comparative analysis demonstrates that YOLOSCEMA achieves superior accuracy-efficiency trade-offs, making it particularly suitable for deployment in low-light conditions, dense scenes, and complex structural environments for autonomous navigation, robotic surveillance, and industrial automation applications.