数据和知识驱动的空战目标集群类型综合识别  

Comprehensive recognization of aerial combat target cluster type driven by data and knowledge

作　　者：张会霞 梁彦[1] 马超雄 汪冕 乔殿峰 ZHANG Huixia;LIANG Yan;MA Chaoxiong;WANG Mian;QIAO Dianfeng(Key Laboratory of Information Fusion Technology,School of Automation,Northwestern Polytechnical University,Xi’an 710129,China;The 20th Research Institute of China Electronic Technology Group Corporation,Xi’an 710068,China)

机构地区：[1]西北工业大学自动化学院信息融合技术教育部重点实验室,西安710129 [2]中国电子科技集团第二十研究所,西安710068

出　　处：《航空学报》2023年第8期217-232,共16页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(61873205)。

摘　　要：目标集群类型识别是体系作战样式下态势认知的关键,然而现有集群识别算法主要依据专家知识人工进行判读,难以满足作战态势快速、准确理解的需求。提出数据和知识驱动下的推理机制,构建分层精细化推理的集群场景识别框架,预识别层检测目标运动过程中的集群的分群/合群,根据设计基于边界检测的密度峰值聚类确定群的划分情况,得到集群的初步识别结果;再识别层中综合分析集群执行任务、运动特性、电磁特性,对集群目标的多源特性进行多元知识约束下的推理网络构建,在此基础上利用现有数据进行推理网络参数学习,进而使推理获得更为准确的集群类型识别结果。该框架综合知识和数据的优势具有从粗到精的集群目标识别能力,利用多特征综合推理机制对目标集群精细化分析,实现集群类型的准确识别。在典型的集群作战活动场景下推理置信度和正确率两项指标均优于现有算法,验证了所提方法的有效性,提高空战目标集群类型识别的置信度和准确率。Identification of cluster types is the key to judging the cognition of combat situation.However,the existing cluster type identification algorithms are mainly based on expert knowledge for manual interpretation,imposing diffi⁃culty in satisfying the needs of rapid and accurate understanding of combat situation.To address this problem,we pro⁃pose a reasoning mechanism driven by data and knowledge,constructing a cluster scene recognition framework for hi⁃erarchical refined reasoning.The pre-recognition layer detects the declustering/clustering of clusters during target movement,and determines the clustering based on the design of boundary detection-based density peaks clustering.Then,according to the division of the cluster,the preliminary identification results of the cluster are obtained.In the reidentification layer,the cluster execution tasks,motion characteristics,and electromagnetic characteristics are com⁃prehensively analyzed and further utilized to construct an inference network under the constraint of multi-knowledge on the multi-source characteristics of the cluster target.Then,the existing data is used to learn the parameters of the in⁃ference network so that it can obtain more accurate cluster type identification results.The framework integrates knowl⁃edge and data to enable coarse to fine cluster target recognition,where the multi-feature comprehensive reasoning mechanism is used to comprehensively identify target clusters.This study realizes the refined identification of the clus⁃ter type,and the two indicators of inference confidence and accuracy are better than the existing algorithms in the typi⁃cal cluster combat scenario,demonstrating the effectiveness of the proposed algorithm and improving the confidence and accuracy of aerial combat target cluster type identification.

关 键 词：态势认知 集群识别 贝叶斯推理 分层推理 群分析 

分 类 号：V19[航空宇航科学与技术—人机与环境工程]