基于短时处理和经验模态分解的地面战场目标被动声识别  

Ground Battlefield Target Passive Acoustic Classification Based on Short-Term Analysis and EMD

作　　者：孙国强[1] 樊新海[1] 张传清[1] SUN Guoqiang FAN Xinhai ZHANG Chuanqing(Dept. of Mechanical Engineering, Academy of Armored Forces Engineering, Beijing 100072, China)

机构地区：[1]装甲兵工程学院机械工程系,北京100072

出　　处：《测试技术学报》2017年第5期434-437,共4页Journal of Test and Measurement Technology

基　　金：武器装备军内科研资助项目(2015ZB21)

摘　　要：针对地面战场目标的被动声识别问题,选取具有代表性的两类坦克、两类履带式装甲车以及卡车作为被动声识别目标对象,以雷声和枪声作为干扰噪声信号,对所采集的声信号进行短时能量分析,得到声信号的短时能量谱,计算短时能量平均值,利用阈值法筛选识别枪声信号,根据经验设置阈值范围;而后,利用经验模态分解(The Empirical Mode Decompo-sition,EMD)方法处理声信号,使其自适应分解得到若干IMF分量,计算IMF分量与原信号能量的比值作为特征值构建特征向量,并利用BP神经网络设计分类器,建立了一种地面目标分级识别方法.研究结果表明:该方法对目标工况适应性强,识别率可达90%以上.In order to identify the ground battlefield target through passive acoustic recognition,selected representative objectives include two kinds of tanks,two kinds of crawler armored vehicle,and truck as the noise acquisition;selected thunder and gunshot as interference noise signal,used short-time energy analysis method to process acoustic signals in order to obtain the short-time energy spectrum of acoustic signal,took the average of the short-time energy,used the threshold method to identify the gunshot signal,set threshold range according to the experience.Used EMD method to decompose the acoustic signal which could obtain several IMF components.The energy ratio of the IMF component to the original signal were used as the eigenvalue to construct the eigenvector.Used BP neural network to design the classifier to establish a method of ground target grading recognition.Research indicates that the method is adaptable to the target condition and classification rate can reach more than 90%.

关 键 词：短时能量 经验模态分解(EMD) 目标识别 神经网络 分级识别 

分 类 号：TN911.72[电子电信—通信与信息系统]