【摘要】 针对现有无人机检测领域存在小目标检测困难以及特征提取效果不佳的问题，提出了一种基于协同注意力模块与双向特征融合结构的无人机小目标检测网络。通过设计协同注意力单元，与骨干网络形成协同注意力模块，显著提升模型特征表达能力；同时借鉴路径聚合网络的思想，设计双向特征融合结构，引入自底向上的特征融合路径，实现浅层特征与深层特征的深度融合，提高小目标检测精度。实验结果表明，所提算法在无人机小目标数据集上的平均精度优于其他网络，显著提升了无人机场景下小目标的检测性能。更多还原

【Abstract】 In view of the difficulties in small target detection and poor feature extraction performance in the existing unmanned aerial vehicle(UAV) detection field, a UAV small target detection network based on collaborative attention module and dual-direction feature fusion(CDNet) was proposed.By designing collaborative attention unit to form collaborative attention module with the backbone network, the feature expression ability of the model was significantly improved.Meanwhile, the dual-direction feature fusion structure was designed based on the idea of path aggregation network, and the bottom-up feature fusion path was introduced to realize the deep fusion of shallow features and deep features so as to improve the detection accuracy of small targets.The experimental results show that the average accuracy of the proposed algorithm on the UAV small target dataset is better than other networks, which significantly improves the detection performance of small targets of UAV.更多还原