论文部分内容阅读
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%。
Aiming at the characteristics of the cockpit voice recorder (CVR), such as the background information of the cockpit recorded in the CVR is complex and complicated, the frequency range is wide and non-stationary, the 15 kinds of cockpit information are Fourier transformed and wavelet packet transformed, Cepstrum coefficient (MFCC) and wavelet packet decomposition coefficient (WPC), the distance separability criterion is used to compress and fuse the MFCC and WPC information to obtain the carcass information eigenvector. A fuzzy support vector machine (FSVM) is designed for unbalanced samples. Two kinds of membership information for each class sample and each kind of carneau information are calculated. Then FSVM classifies and identifies the carcasses, and then resolves the CVR signal The results show that this method is obviously superior to the conventional support vector machines (SVM) and FSVM, and the recognition rate of classification is up to 98.33%.