基于深度学习的目标行为知识发现方法  

Method of Target-action Knowledge Discovery Based on Deep Learning

作　　者：成磊峰 何丽莎 薛丽惠 刘欣 CHENG Leifeng;HE Lisha;XUE Lihui;LIU Xin(Southwest China Institute of Electronic Technology,Chengdu 610036)

机构地区：[1]中国西南电子技术研究所,成都610036

出　　处：《计算机与数字工程》2022年第3期532-537,共6页Computer & Digital Engineering

摘　　要：针对情报处理领域的知识动态更新问题,提出了一种基于深度学习的目标行为知识发现方法。首先,根据已有知识进行目标数据预处理,实现对目标要素补充;其次,基于已掌握行为意图的历史数据,构建目标行为特征模型;然后,基于深度学习方法实现行为规律分析,为目标识别提供知识支撑;最后,通过目标识别及识别结果的变化异常分析进行新知识发现,形成知识使用、发现的良性循环。通过数值仿真,该方法对舰机目标识别准确率在80%左右,且能有效发现新知识。Aiming at the problem of knowledge dynamic updating in the field of information processing,a method of target-action knowledge discovery based on deep learning is proposed.First of all,the target data is preprocessed according to the existing knowledge to supplement the target elements.Secondly,based on the historical data of behavior intention,the target-action feature model is constructed.Thirdly,based on the deep learning method,the target-action law is analyzed to provide knowledge support for target recognition.Finally,new knowledge discovery is carried out through target recognition and abnormal analysis of recognition results,forming a virtuous circle of knowledge use and discovery.Through numerical simulation,the accuracy of this method is about 80%,and new knowledge can be found effectively.

关 键 词：情报处理 深度学习 目标行为 知识发现 

分 类 号：TN957.52[电子电信—信号与信息处理]