基于自动编码器和支持向量机的飞机机动智能识别方法  

Aircraft maneuver recognition based on autoencoder-supportvector machine

作　　者：岳龙飞 杨任农[2] 杨文达 左家亮[2] 刘会亮 许凌凯 YUE Longfei;YANG Rennong;YANG Wenda;ZUO Jialiang;LIU Huiliang;XU Lingkai(Naval University of Engineering,National Key Laboratory of Electromagnetic Energy,Wuhan 430033,China;Air Traffic Control and Navigation College,Air Force Engineering University,Xi’an 710051,China;Xi’an Satellite Monitoring and Control Center,Xi’an 710043,China;Unit 95882 of the Chinese People’s Liberation Army)

机构地区：[1]海军工程大学电磁能技术全国重点实验室,武汉430033 [2]空军工程大学空管领航学院,西安710051 [3]西安卫星测控中心,西安710043 [4]中国人民解放军95882部队

出　　处：《兵器装备工程学报》2024年第2期210-217,共8页Journal of Ordnance Equipment Engineering

基　　金：国家自然科学基金青年项目(62106284);陕西省自然科学基金项目(2021JQ-370)。

摘　　要：飞机机动动作识别是空战意图识别和智能决策的基础。针对传统机动动作识别方法存在的高维数据分析和特征提取能力不足、识别准确率不高的问题,考虑到机动数据的高维性、时序性的特点,提出基于正则化自动编码器-支持向量机(RAE-SVM)的飞机机动动作识别方法。依据机动动作数据变化规律和专家经验知识,构建了基于时间段数据特征的机动动作样本库;将无监督的自动编码器神经网络强大的特征提取能力和有监督的支持向量机优异的分类性能相结合,构建基于RAE-SVM的机动识别模型,采用机动动作样本库训练模型;通过引入正则化提高了RAE网络的泛化性能和预测准确率;最后与多种现有方法进行准确性与实时性对比,并选取空战机动动作数据进行实例验证。结果表明:所提方法识别准确率为92.75%,对一组机动数据识别仅需2 ms,满足实时性要求。因此,该方法可以快速准确地进行飞机机动动作识别,具有一定实用价值。Aircraft maneuver recognition is a foundation in air combat intention recognition and intelligent decision-making.Aiming at the problems of high dimensional data processing and feature extraction ability and low recognition accuracy of traditional maneuver recognition methods,in view of the high-dimensionality and time-series characteristics of maneuver data,a novel regularized auto-encoder-support vector machines(RAE-SVM)based method is proposed.According to the change law of maneuvering action data and the expert prior knowledge,the maneuver recognition sample library based on time period data features is constructed.Combining powerful feature extraction capability of unsupervised auto-encoder with superior classification performance of supervised support vector machine,the aircraft maneuver recognition model based on RAE-SVM is constructed and verified by the maneuver recognition sample library.The generalization performance and the accuracy of RAE network are improved by introducing regularization.The simulation results show that the recognition accuracy of the proposed method is 92.75%.The trained model takes 2 milliseconds to recognize a set of maneuver data and the running time meets the real-time requirements.Therefore,the proposed method can quickly and accurately recognize aircraft maneuver,which has certain practical value.

关 键 词：机动动作识别 正则化自动编码器 支持向量机 特征提取 无监督学习 

分 类 号：E844[军事—战术学]