Anti-Drone Detection in Aerial Multispectral Images

Anti-drone detection is becoming a crucial application of remote sensing technology due to the rising security risks associated with drones. Traditional RGB cameras face challenges in detecting drones in dim environments, such as vegetation and marine scenarios. While multispectral sensors incorporating the red-edge and near-infrared spectra have shown advantages over common RGB sensors in agriculture and human rescue applications, this study explores their new application to anti-drone detection. To accommodate multispectral modalities, we introduce a multi-modal deep learning method termed the multispectral-fusion network (MSF-Net), which features a dual-flow structure and a multi-modal fusion (MMF) module. We construct a drone-borne multispectral sensor to track a target drone, capturing its aerial images and establishing a dataset with 3,000 samples. Experimental results have shown that the dual-flow MSF-Net significantly outperforms both the RGB-only and single-flow multispectral methods in anti-drone detection performance. Moreover, the MMF module notably enhances the detection of small drones.