基于注意力机制和知识蒸馏的跨模态遮蔽目标检测OA

Because of the concealment feature,surface covered targets,e.g.,underground caverns and aircraft shelters,will cause problems such as background noise interference and lack of feature information,making them impossible to be accurately detected.To address this issue,in this paper,a cross-modal object detection network based on the attention mechanism and knowledge distillation is proposed.A feature enhanced module based on the attention mechanism is designed in the backbone network to improve the ability of the network to extract tiny features of the target.The overall network is an optical-synthetic aperture radar(SAR)dual-branch network,and a cross-modal knowledge distillation module is designed to transfer the feature information of the optical branch network and improve the target detection performance in SAR images.In addition,a paired optical-SAR surface covered target detection dataset is constructed,including 1 032 pairs of underground hangar images and 682 pairs of underground cave images.The experimental results on the dataset show that the proposed method is better than other traditional algorithms,achieving an mAP50 of 96.1%,which has significant improvement compared with the traditional single-modal and existing cross-modal detection algorithms,proving the effectiveness of the proposed algorithm in surface covered target detection.