基于倒谱特征和小波包特征熵的直升机声目标识别  

The Recognition of Helicopter Acoustic Target Based on Cepstrum Characteristic and Wavelet Packet Characteristic Entropy

作　　者：黄博[1] 高勇[1] 

机构地区：[1]四川大学电子信息学院,四川成都610065

出　　处：《探测与控制学报》2007年第6期15-18,23,共5页Journal of Detection & Control

基　　金：总装预研基金项目资助(基金号不公开)

摘　　要：提出了一种将倒谱特征和小波包特征熵相结合的直升机声目标识别新算法,首先分析了直升机声信号的特点,计算了声信号的MFCC(MEL频率倒谱系数)、差分MFCC(差分MEL频率倒谱系数)和小波包分解后各个频带内的小波包特征熵组成的特征向量,并以此向量输入反向误差传播(Back Propagation,BP)神经网络进行训练,再用训练好的神经网络进行不同直升机型号的识别,最后给出了统计结果。结果表明:该算法对直升机机型的识别有较好的效果。A new algorithm of the recognition of helicopter acoustic signal is brought forward by combining cepstrum characteristic and wavelet packet characteristic entropy. First of all the trait of helicopter acoustic signal is analyzed. The characteristic vectors are composed of computed MFCC （MEL frequency cepstrum coefficient） and difference MFCC （difference MEL frequency cepstrum coefficient） as well as wavelet packet characteristic entropy in every frequency band after wavelet packet is decomposed. The vectors are input in a BP （Back Propagation） neural network in order to train the network, and then the trained BP network is used to detect the different models of helicopters. The statistical result is given in the end. The results indicate the algorithm has better effect for the recognition of different kinds of helicopters.

关 键 词：识别 直升机声信号 MEL倒谱系数 小波包特征熵 特征向量 BP神经网络 

分 类 号：V275.1[航空宇航科学与技术—飞行器设计] TP391.42[自动化与计算机技术—计算机应用技术]