基于EEMD-SVM非平衡决策树的人员及车辆识别方法  

Method of People and Vehicles Identification Based on EEMD-SVM Unbalance Decision Tree

作　　者：赵春晖[1] 王杨[1] 唐胜武[1,2] 王祎帆 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001 [2]中国电子科技集团公司第四十九研究所,黑龙江哈尔滨150001

出　　处：《沈阳大学学报（自然科学版）》2016年第5期384-391,共8页Journal of Shenyang University：Natural Science

基　　金：国家自然科学基金资助项目(61571145;61405041)

摘　　要：通过对人员、轮式车、履带车产生的地震动信号进行分析,建立人员及车辆识别系统模型.针对人员及车辆产生的地震动信号的非线性和非平稳特征,采用集成经验模态分解(ensemble empirical mode decomposition,EEMD)算法对实测人员、车辆产生的地震动信号进行分解,然后对分解得到的固有模态函数(intrinsic mode function,IMF)高频分量进行小波阈值去噪.选取有效的IMF分量,计算其归一化能量特征矩阵.再将特征矩阵输入到支持向量机(support vector machine,SVM)非平衡决策树分类器中,进行人员、轮式车和履带车的逐层识别.实验结果表明,EEMD-SVM非平衡决策树模型可以准确、高效地对人员、轮式车和履带车进行分类识别.Through analyzing the vibration signal of people,wheeled vehicles and tracked vehicles,a people and vehicles identification system model was established.Aiming at the nonlinear and non-stationary characteristics,EEMD （ensemble empirical mode decomposition）was used to decompose people and vehicles vibration signal to obtain a number of IMFs（intrinsic mode function）.Then,the high frequency components of the IMF were used wavelet threshold algorithm for denoising.Selecting a valid component of IMFs and calculating normalized energy matrix as the input of SVM（support vector machine）unbalance decision tree classifier,people,wheeled vehicles and tracked vehicles were classified layer by layer through the SVM.The experimental results demonstrate the the superior performance of EEMD-SVM unbalance decision tree mode in people,wheeled vehicles and tracked vehicles classification.

关 键 词：集成经验模态分解 固有模态函数 能量特征矩阵 SVM非平衡决策树分类器 

分 类 号：TN911.6[电子电信—通信与信息系统] TN911.7[电子电信—信息与通信工程]