The rapid proliferation of unmanned aerial vehicles (UAVs) has amplified the need for robust and efficient object detection in diverse aerial environments. However, detecting small objects under complex conditions (e.g., low illumination, cluttered backgrounds, and thermal–visual discrepancies) remains challenging. While many existing detectors emphasize real-time inference, they often rely on weak or late fusion strategies, resulting in suboptimal utilization of complementary multi-modal cues. To address this limitation, we propose DGE-YOLO, an enhanced YOLO-based framework for effective infrared–visible (IR–RGB) multi-modal fusion in UAV object detection. DGE-YOLO adopts a dual-branch architecture for modality-specific feature extraction, preserving modality-aware representations before fusion. To strengthen cross-scale semantics, we introduce an Efficient Multi-scale Attention (EMA) module that improves feature discrimination across spatial resolutions. Furthermore, we replace the conventional neck with a Gather-and-Distribute module to reduce information loss during feature aggregation and improve multi-scale feature propagation. Extensive experiments on the DroneVehicle dataset demonstrate that DGE-YOLO consistently outperforms state-of-the-art baselines, confirming its effectiveness and practicality as an applied multi-modal detection solution for UAV scenarios.
Traditional fall detection methods face significant limitations in complex environments, particularly under occlusion and poor lighting conditions. To address these challenges and enhance the detection accuracy in intelligent real-time monitoring systems, this study proposes an optimized BMR-YOLO framework based on the YOLOv8n architecture. Our approach introduces four strategic improvements to effectively overcome environmental constraints. First, we enhance the backbone network by incorporating a BiFormer vision transformer with dual-layer routing attention, enabling dynamic allocation of computational resources, while improving both computational efficiency and feature extraction performance. Next, we replace the C2f module in the backbone with C2f_rvb, enhancing the model’s ability to handle multi-scale features while reducing computational requirements. Furthermore, we strengthen the detection head by adding the MultiSEAM attention mechanism, significantly improving the detection accuracy of occluded objects. Finally, we optimize the regression process by replacing the traditional CIoU loss with a direction-aware SIoU metric, thereby improving both the localization accuracy and training stability of bounding box regression. To validate our method, we constructed a comprehensive dataset, BMR-fall, containing over 10,000 annotated images that capture various fall scenarios, and performed cross-validation using the UR fall detection dataset. Experimental results demonstrate that BMR-YOLO achieves a notable improvement in mAP@0.5, rising from 0.852 to 0.899 on our proprietary dataset, while maintaining a low computational cost of 6.5 GFLOPs. Comparative analysis with existing methods shows that BMR-YOLO outperforms them under occlusion and lighting variation conditions, confirming the model’s robustness and practical applicability for real-world deployment.