基于LMD局部投影能量特征的车型识别

发布时间：2018-05-08 12:17

本文选题：交通工程 + 车型识别　；参考：《中国公路学报》2017年01期

【摘要】：针对动态噪声环境下行进中的机动车辐射出的声信号具有强非平稳性、多尺度性及低信噪比的问题,提出一种基于局部均值分解(LMD)和局部投影能量计算的车型声特征提取方法。首先,利用LMD方法对采集的声信号进行自适应分解,得到各尺度上的乘积函数(PF)分量,从强背景噪声中分离出包含车型特征频率成分的PF分量;其次,对LMD分解结果进行加权优化,重构特征PF分量,滤除虚假成分及弱相关分量,以增强特征信息;最后,将特征PF分量的能量等距离投影到能量聚集区内,基于能量尺度构造声信号的低维特征向量,并通过人工神经网络的学习对特征向量进行识别。在试验数据集上,采用LMD局部投影能量特征对目标车辆进行车型识别,并对试验数据集添加不同强度的噪声,进行LMD分解及局部投影能量计算,将计算结果与其他特征提取方法计算结果进行对比分析。结果表明:该方法对于车型信息十分敏感,识别率达到93.4%;可以有效抑制动态环境下的背景噪声干扰,获取目标敏感的窄带信号,具有很好的抗噪能力;选择在重构窄带信号的能量聚集区内进行投影计算,可以有效去除冗余特征,同时提高算法的实时性。
[Abstract]:In order to solve the problems of strong non-stationary, multi-scale and low signal-to-noise ratio (SNR) of the acoustic signals emitted by moving vehicles in dynamic noise environment, a method of vehicle acoustic feature extraction based on local mean decomposition (LMD) and local projection energy calculation is proposed. Firstly, the LMD method is used to decompose the collected acoustic signal adaptively, and then the product function (PF) component is obtained at each scale, and the PF component, which contains the characteristic frequency component of the vehicle model, is separated from the strong background noise. The LMD decomposition results are weighted and optimized to reconstruct the feature PF components, filter out the false components and the weak correlation components to enhance the feature information. Finally, the energy equidistant of the feature PF components is projected into the energy accumulation area. The low dimensional eigenvector of acoustic signal is constructed based on energy scale, and the feature vector is recognized by learning from artificial neural network. On the test data set, the vehicle model is identified by using the LMD local projection energy feature, and the noise of different intensity is added to the test data set, and the LMD decomposition and local projection energy calculation are carried out. The calculated results are compared with those of other feature extraction methods. The results show that the method is very sensitive to vehicle information, and the recognition rate is 93.4.The method can effectively suppress background noise interference in dynamic environment, obtain narrow band signal sensitive to target, and has good anti-noise ability. The projection calculation in the energy accumulation area of the reconstructed narrow band signal can effectively remove the redundant features and improve the real-time performance of the algorithm.
【作者单位】：天津大学电气与自动化工程学院;沈阳大学信息工程学院;
【基金】：国家自然科学基金项目(61473202) 辽宁省自然科学基金项目(201602520) 沈阳市科技计划项目(F15-126-9-00)
【分类号】：U495

【相似文献】