低空飞行目标声音优化识别研究

Research on Optimization of Low Altitude Acoustic Flight Target Recognition

针对战场环境噪声下的低空目标声音识别问题,首先用阈值EMD对典型环境噪声进行去除,其次提取基于离散小波变换的Mel频率倒谱系数(WMFCC)的特征参数,最后利用支持向量机(SVM)分类识别,利用含噪声音提取的MFCC和WMFCC以及经过阈值EMD去噪后的声音再提取的MFCC和WMFCC作为特征向量进行分类识别。不同噪声不同信噪比条件下的对比实验结果表明,SVM分类器利用阈值EMD去噪后提取的WMFCC特征参数进行分类,可以有效去除噪声,提高识别率,并且在低信噪比环境噪声下,分类性能明显优于其它方法。

In order to solve the problem about the acoustic recognition of low altitude target under battlefield ambient noise,we firstly used threshold EMD to remove typical ambient noise and extracted the feature parameter of Mel frequency cepstral coefficients based on discrete wavelet transform（ WMFCC）. Then,we used support vector machine（ SVM） to recognize and classify it. Thus,we extracted MFCC and WMFCC from the noisy sound,and extracted MFCC and WMFCC by the sound after EMD denoising as feature vectors for classification and recognition. Following conclusion can be drawn from contrast experimental results based on different noises and different signal-to-noise ratios. When using WMFCC feature parameter extracted after denosing threshold EMD for the classification,SVM classifier can effectively remove noise and improve recognition rate. Meanwhile,the classification performance is significantly better than other methods in the low signal to noise ratio environment.