基于时空推理网络的空中红外目标抗干扰识别算法  

Anti-interference recognition method of aerial infrared targets based on a spatio-temporal correlation inference network

作　　者：张良 田晓倩 李少毅[3] 杨曦 Zhang Liang;Tian Xiaoqian;Li Shaoyi;Yang Xi(China Airborne Missile Academy,Luoyang 471009,China;Aviation Key Laboratory of Science and Technology on Airborne Guided Weapon,Luoyang 471009,China;School of Astronautics,Northwestern Polytechnical University,Xi’an 710072,China)

机构地区：[1]中国空空导弹研究院,河南洛阳471009 [2]航空制导武器航空科技重点实验室,河南洛阳471009 [3]西北工业大学航天学院,陕西西安710072

出　　处：《红外与激光工程》2022年第7期463-472,共10页Infrared and Laser Engineering

基　　金：国家自然科学基金(61703337);航空基金(ASFC20191053002)。

摘　　要：复杂空战背景下的抗红外诱饵干扰技术是红外空空导弹的核心技术之一。针对传统静态贝叶斯网络不能表达序列图像中特征变量在时序上动态变化关系,提出了一种符合人类视觉推理识别过程的时空关联推理网络抗干扰识别算法。首先,提出的时空关联推理网络在考虑特征空间约束关系的基础上,引入了特征变量时间约束的先验知识,建立表达特征时空关联的目标推理网络识别模型,从而增强了序列图像目标识别的稳定性;其次,通过仿真数据构建样本集,离线训练学习时空关联推理网络结构及特征跳转概率参数,确定概率推理网络识别离线模型;最后,依据测试数据,结合推理识别网络模型进行概率推理,实现对目标的识别分类。实验结果表明,在伴随红外诱饵干扰投放的条件下,基于时空关联推理网络的抗干扰识别率达到94%,比静态贝叶斯网络抗干扰识别算法高3%,有效提升了目标识别的稳定性。The infrared anti-interference technique of missiles under the background of complex air combat is one of the core technologies of infrared air-to-air missiles.Aiming at the fact that traditional static Bayesian networks cannot express the dynamic relationship of feature variables in sequence images in time series,this paper proposes an anti-jamming recognition algorithm for a space-time correlation inference network that conforms to the process of human visual inference and recognition.First,the proposed space-time association reasoning network takes into account the feature space constraint relationship,introduces prior knowledge of the time constraints of feature variables,and establishes a target reasoning network recognition model that expresses the characteristic spatiotemporal relationship,thereby enhancing the stability of sequence image target recognition.Second,a sample set is built through simulation data,offline training and learning the space-time correlation inference network structure and feature jump probability parameters,to determine the probabilistic inference network to identify the offline model.Finally,based on the test data,the model is combined with the inference identification network model to perform probabilistic inference to achieve recognition and classification of targets.The experimental results show that the anti-jamming recognition rate based on the spatiotemporal correlation inference network reaches 94% under the condition of the interference of the infrared decoy,which is3% higher than the static Bayesian network anti-jamming recognition algorithm,which effectively improves the stability of target recognition.

关 键 词：红外空空导弹 抗干扰 目标识别 时空关联 推理网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]