面向不均匀飞机舱音信号样本的模糊支持向量机识别方法  

Signal Recognition of Imbalanced Samples for CVR Based on Fuzzy SVM

作　　者：杨琳 王从庆[2] 姜龙生[2] 

机构地区：[1]中国民航科学技术研究院,北京100028 [2]南京航空航天大学自动化学院,江苏南京210016

出　　处：《航空学报》2012年第3期544-553,共10页Acta Aeronautica et Astronautica Sinica

基　　金：国家自然科学基金(60776819)~~

摘　　要：针对驾驶舱话音记录器(CVR)中记录的舱音背景信息多而复杂、频率范围宽、非平稳等特点,通过对15种舱音信息进行傅里叶变换和小波包变换,依次提取其Mel倒谱系数(MFCC)和小波包分解系数(WPC),利用距离可分性判据对MFCC和WPC信息进行压缩融合,得到舱音信息特征向量。设计了面向不均衡样本的模糊支持向量机(FSVM),分别计算每种类别样本及其内每种舱音信息的2个隶属度,然后利用FSVM对舱音信号进行分类识别,解决了CVR信号含噪奇异样本和数目不均衡样本时识别性能较差的缺点,实验表明该方法明显优于常规支持向量机(SVM)和FSVM,分类识别率达到98.33%。The voice signals in a cockpit voice recorder（CVR）are complex,non-stationary,and exist in a wide frequency range.By using Fourier transform and wavelet packet transform for the fifteen kinds of signals in a CVR,Mel frequency cepstrum coefficient（MFCC） and wavelet packet coefficient（WPC） are extracted as the initial characteristic samples.The characteristic vectors are determined by compression of the MFCC and WPC samples using a geometric distance classification criterion.A fuzzy support vector machine（FSVM） is designed to handle the imbalanced sample classification in the CVR,in which two different fuzzy-membership values in relation to the imbalanced samples are calculated by the extracted samples in each voice signal.The above method can improve the recognition performance of the voice signals with imbalanced samples in the presence of outliers and noise.The experimental results show that it is obviously superior to the conventional support vector machine（SVM） and FSVM with a 98.33% recognition rate.

关 键 词：驾驶舱话音记录器 非话语背景声音信号 特征提取 不均匀舱音样本 模糊支持向量机 

分 类 号：V248.2[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]