Advanced drone detection via ground-based RGB surveillance using efficient deep learning architectures

The rapid proliferation of unmanned aerial vehicles (UAVs) in civilian airspace has introduced serious security challenges. Unauthorized drones can disrupt airport operations and may also be exploited for illicit activities, creating an urgent need for reliable detection systems. Sensor-based counter-drone solutions such as radar, radio frequency, acoustic, and thermal sensing have been widely studied, but each modality can degrade under specific conditions. Vision-based detection with standard RGB cameras is attractive due to low cost and rich visual cues. However, detecting small drones in complex outdoor scenes is difficult because distant micro-drones occupy only a few pixels and are often confused with birds or background clutter, causing missed detections and false alarms. This paper proposes the Hybrid Efficient Drone Detector (HEDD), an efficient deep learning framework for robust small-drone detection in RGB video. HEDD introduces an edge-aware feature extraction module that enhances object contours from single-camera input, combined with a lightweight convolutional backbone and multi-scale feature fusion. This design improves sensitivity to tiny targets across distances while preserving real-time throughput. Experiments on the DUT Anti-UAV benchmark and a custom surveillance dataset show that HEDD achieves high accuracy, low false-alarm rates, and competitive or superior performance compared with state-of-the-art methods.