基于并行注意力机制的地面红外目标检测方法(特邀)

Ground infrared target detection method based on a parallel attention mechanism(Invited)

地面背景下的红外目标检测是伪装防护、精确制导等领域的关键技术。针对现有基于深度学习的目标检测模型对地面背景下红外目标进行检测时容易受到复杂背景干扰、对目标关注不足,从而导致检测准确率不高的问题,文中提出了一种基于并行注意力机制的地面红外目标检测方法。首先,利用卷积和注意力并行的下采样方式,在降低模型的空间复杂度和提升训练速度的同时,对目标特征进行聚焦和关注;其次,对主干网络提取的多尺度特征进行融合,通过不同尺度信息的复用与互补抑制背景信息的干扰,提升目标检测的准确率;最后,利用焦点损失函数和CIOU损失函数提高模型的分类与回归精度。实验结果表明,在Infrared-VOC数据集上该模型的平均检测精度为82.2%,比YOLOv3提高了6.9%,同时模型的空间复杂度仅为YOLOv3的32.6%,训练时间为YOLOv3的43.7%,实现了模型训练效率和检测精度的提升。

Ground infrared target detection is a key technology in the fields of camouflage protection and precision guidance.For the current deep learning-based target detection model to detect infrared targets in the ground background,it is easy to be interfered by complex backgrounds and insufficient attention to the target,which leads to the problem of low detection accuracy.A method of ground infrared target detection based on a parallel attention mechanism was proposed.Firstly,the parallel down-sampling method of convolution and attention was used to reduce the spatial complexity of the model and increase the training speed,while focusing and paying attention to the target features.Secondly,the multi-scale features extracted by the backbone network were fused to suppress the interference of background information and improve the accuracy of target detection through the multiplexing and complementary of different scale information.Finally,the focal loss and CIOU loss were used to improve the classification and regression accuracy of the model.The experiment results showed that the average detection accuracy of the model on the Infrared-VOC dataset was 82.2%,which was 6.9%higher than YOLOv3.At the same time,the space complexity of the model was only 32.6%of YOLOv3,and the training time was 43.7%of YOLOv3.The improvement of model training efficiency and detection accuracy was achieved.