Abstract: Air-to-ground target detection plays an important role in performing typical military and civilian tasks on air-based unmanned platforms such as drones and missiles.Aiming at the problems of fewer effective features and high false detection rate of the air-to-ground target intelligent detection method,an air-to-ground target intelligent detection algorithm based on the dual attention mechanism is proposed.First,a channel attention mechanism is established,focusing on the important channels in the feature layer that contain the effective features of the target.Second,the spatial attention mechanism is used to enable the network to focus on local key target areas.The dual attention mechanism combines the advantages of the channel attention mechanism and the spatial attention mechanism to improve the efficiency of target detection.A loss function that couples the dual attention mechanism and target detection is also designed,which realizes the synchronous optimization of the loss function.Finally,a comparative experiment is performed using the air-to-ground target detection data set.The experimental results show that the proposed algorithm has higher target detection accuracy and training speed than other algorithms in the air-to-ground scenario.

Key words: air-to-ground target detection, deep learning, complex feature interference, dual attention mechanism, effective feature, key area

摘要： 空对地目标检测在无人机、导弹等空基无人平台执行典型军事和民用任务中均发挥着重要作用。针对空对地目标检测存在目标有效特征少和误检测率高等问题,提出了一种基于双重注意机制的空对地目标智能检测算法。首先,建立通道注意机制,重点关注特征层中包含目标有效特征的重要通道;其次,利用空间注意机制,关注局部重点目标区域;而双重注意机制结合了通道注意机制和空间注意机制的优势,提升目标检测效率。还设计了双重注意机制和目标检测两者相耦合的损失函数,实现了损失函数的同步优化;最后,利用空对地目标检测数据集进行了对比实验。实验结果表明,在空对地场景下,所提出的算法与其他算法相比,具有更高目标检测精度和训练速度。

关键词: 空对地目标检测, 深度学习, 复杂特征干扰, 双重注意机制, 有效特征, 重点区域