Skip to content
Conference Open access

RCDNet: A Detection Model for Rotary-Wing UAVs

Sep 2026 · 2026 7th International Conference on Intelligent Computing and Human-Computer Interaction (ICHCI) · pp. 267-270 · 0 citations · 5 references

Abstract

The widespread adoption of rotary-wing UAVs in aerial surveying, precision agriculture, and logistics delivery has posed significant threats to urban public safety and airspace order due to their low-altitude, small-sized, and slow-moving characteristics, where rapid and precise detection serves as the prerequisite for effective supervision and defense. To address the challenges of existing algorithms in UAV detection, including low accuracy in small target recognition, insufficient multi-scale feature fusion, and redundant model parameters, this paper proposes RCDNet, a UAV detection algorithm based on YOLOv11. First, the RGCA module is designed in the backbone network, which reduces redundant computation through channel splitting and hierarchical convolution, introduces RepConv to ensure efficient gradient flow while discarding residual structures, and leverages structural reparameterization to achieve training-inference decoupling, enhancing multi-scale feature extraction at zero extra inference cost. Second, the CGFM module is proposed in the neck network, replacing the original simple concatenation with a bidirectional channel attention mechanism to achieve adaptive weighted fusion of shallow detail and deep semantic features, effectively strengthening rotor edge details while suppressing background interference. Finally, the SDEH is designed as a shared detail-enhanced detection head, which compresses parameters through cross-scale weight sharing, embeds detail-enhanced convolution with multi-directional differential priors to strengthen small-target edge responses, and ensures inference efficiency via structural reparameterization. Experimental results demonstrate that RCDNet achieves 86.9% mAP@0.5 and 57.2% mAP@0.5:0.95, improving by 2.9% and 2.8% over the baseline YOLOv11n, while reducing parameters to 2.27 M and model size to 5.0 MB, effectively balancing detection accuracy and model complexity, thus providing an efficient and feasible solution for accurate rotary-wing UAV detection.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.