基于DLPCC和ELM的装甲车辆声识别  

Research on Armored Vehicle Acoustic Classification Based on DLPCC and ELM

作　　者：樊新海 石文雷 张传清 FAN Xinhai;SHI Wenlei;ZHANG Chuanqing(Army Academy of Armored Forces,Vehicle Engineering Department,Beijing 100072,China)

机构地区：[1]陆军装甲兵学院车辆工程系,北京100072

出　　处：《兵器装备工程学报》2018年第7期20-25,共6页Journal of Ordnance Equipment Engineering

基　　金：武器装备军内科研项目(2015ZB21)

摘　　要：以常见的3种坦克和4种履带式装甲车为识别对象,提出了一种基于动态线性预测倒谱系数(DLPCC)和极限学习机(ELM)的装甲车辆声识别模型。该模型以DLPCC为特征,对不同工况下的装甲车辆噪声信号进行特征提取。根据噪声信号特征对ELM进行特征训练和识别,获得噪声识别模型。实验结果表明,DLPCC能更好地体现噪声特征,识别效果优于传统的LPCC。与传统的BP神经网络以及概率神经网络(PNN)相比,以极限学习机为分类器的识别模型具有用时短、准确率高的特点,识别率达到91%以上。A kind of sound recognition classifier model of armored vehicle based on Dynamic Linear Prediction Cepstral Coefficients （DLPCC） taken as the feature extraction method and the Extreme Learning Machine （ELM） was established, in which three kind of tanks and four kinds of tracked armored vehicles were taken as the recognition object. The univariate analysis was used to analyze the core parameters that affects ELM recognition rates, and the best acoustic recognition model was obtained. The experimental results show that DLPCC can better reflect the characteristics of noise, and the recognition effect is better than that of the traditional LPCC. Compared with the traditional BP neural network and probabilistic neural network （PNN）, the recognition model based on ELM has the characteristics of short time and high accuracy, and the recognition rate is over 91%.

关 键 词：动态线性预测倒谱系数 极限学习机 特征提取 声识别 

分 类 号：TJ811[兵器科学与技术—武器系统与运用工程]