Answers are grounded in this paper and cite their sources.
Nighttime vehicle detection using unmanned aerial vehicles (UAVs) is important for traffic monitoring, road safety management, and emergency response. However, substantial variations in object scale, low target-to-background contrast, complex backgrounds, and limited onboard computing resources pose considerable challenges to accurate and efficient detection. To address these challenges, this study proposes SGL-YOLO (Similarity-guided Global-Lightweight YOLO), a lightweight detector based on YOLO26. We propose a similarity-based attention bidirectional feature pyramid network (SBA-FPN) that integrates bidirectional feature propagation, adaptive weighted fusion, and parameter-free attention to enhance multiscale feature representation and preserve informative vehicle features across different scales. We propose a global context enhancement module (GCEM) that jointly models channel and spatial importance to strengthen vehicle-related responses while suppressing interference from complex backgrounds. A Ghost-DFC Fusion Block (GDFB) is proposed to combine lightweight local feature extraction with long-range spatial modeling, enabling effective integration of vehicle structural features and contextual information with limited computational overhead. Experiments show that SGL-YOLO achieves mean average precision (mAP) of 57.3% on DroneVehicle Night and 43.6% on VisDrone Night, outperforming all selected comparison models on both datasets. These results demonstrate the effectiveness of the architecture for vehicle detection in complex nighttime UAV imagery and highlight its potential for resource-constrained UAV traffic monitoring.
No takes yet. Share an insight, caveat, or question.
Song et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: