ECE-YOLO: An Efficient Detection Model for UAV Remote Sensing Object Detection

To address the challenges of drastic scale variations, dense distribution of small objects, and complex background interference in UAV remote sensing object detection, as well as the high resource consumption of existing models under high-accuracy settings, this paper proposes a more efficient object detection method termed ECE-YOLO. First, an Enhanced Spatial Perception Stem (ESPMStem) is designed, which introduces an importance-aware spatial modulation mechanism before the downsampling operation to strengthen shallow spatial perception capability and improve small-object detection performance. Second, a Context-Modulated Re-parameterized Module (CM-RepNCSPELAN4) is proposed to replace the C3K2 module. This module incorporates the Efficient Channel Attention (ECA) mechanism, a Global Context (GC) modeling module, and a dynamic channel modulation mechanism, thereby enhancing feature representation and category discrimination capabilities for objects at different scales. Finally, a lightweight feature fusion structure with unified output channels (UCF-Neck) is proposed to reduce model complexity while better adapting to the dynamic detection head. In addition, an ECA-enhanced Dynamic Head (ECA-DyHead) is designed to address the limited feature selection capability of dynamic heads along the channel dimension, thereby improving multi-scale object adaptability and suppressing false detections. Experimental results on the VisDrone2019-DET dataset demonstrate that the proposed method improves mAP@50 from 32.8% to 36.6% (+3.8%) and mAP@50–95 from 19.3% to 21.9% (+2.6%), showing significant performance gains. Meanwhile, the number of model parameters is reduced from 2.59M to 2.10M, effectively lowering model complexity while improving detection accuracy. Stable performance improvements are also achieved on the UAVDT dataset, validating the good generalization capability of the proposed method. The results indicate that the proposed approach achieves a favorable balance between detection accuracy and computational efficiency in complex UAV remote sensing scenarios.