近地背景下的机载红外探测系统弱小目标检测  

Weak and small targets detection of airborne infrared detection system in the Near-earth background

作　　者：李楠 LI Nan(Modern Logistics College,Shanxi Vocational University of Engineering Science and Technology,Jinzhong 030600,China)

机构地区：[1]山西工程科技职业大学现代物流学院,山西晋中030600

出　　处：《光学技术》2022年第6期755-762,共8页Optical Technique

摘　　要：机载红外探测系统在近地背景下检测目标时,地面将对弱小目标产生严重的干扰,导致传统检测方法对弱小目标的检测性能下降。针对该问题,利用生成对抗网络提出一种近地背景下的机载红外探测系统弱小目标检测方法。将深度自编码器作为生成对抗网络的网络框架,引入inception机制对视觉信息进行多尺度特征提取,并引入残差块来缓解梯度消失问题。在神经网络的对抗训练中,生成器考虑了移动损失与对抗损失两个损失函数,提高了生成器的训练效果。最终,在公开的无人机机载红外探测数据集上完成了实验,结果表明所提方法能在近地背景下成功检测出红外弱小目标,且检测的平均精度与速率均优于其它对比方法。When airborne infrared detection system detects targets in the near-earth background,the ground will disturb the weak and small targets seriously,which leads to detection performance reduction of traditional detection methods for weak and small targets.In view of this problem,the generative adversarial networks are utilized to propose a weak and small target detection method of airborne infrared detection system in the near-earth background.The generative adversarial networks are constructed based on deep auto-encoder,and the inception mechanism is introduced to extract multiple scaled features of vision information,the residual block is also introduced to mitigate the vanishing gradient issue.In the adversarial training of the neural networks,the loss function of the generator consists of the moving loss and the adversarial loss,which improves the training effect of the generator.In the end,the experiments are carried on the public unmanned drone airborne infrared detection datasets,and the results show that the proposed method can detect the infrared weak and small targets in the near-earth background successfully,at the same time,the average precision and speed of the detection method are better than the compared methods.

关 键 词：机载红外探测系统 户外探测 弱小目标检测 深度自编码器 生成对抗网络 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]