针对战场环境存在噪声干扰的情况,提出了一种基于小波包分析的声目标特征参数提取方法。该方法将小波包分析和Mel倒谱分析相结合,提高了特征参数的鲁棒性。实验结果表明,在噪声条件下,基于小波包分析的平均识别率比MFCC参数提高6.78%,在信噪比为5dB时,识别率仍能达到94.5%。

Aimed at noise in battlefield, a method was proposed for feature extraction from acoustic targets based on wavelet packet analysis. The feature parameter combined wavelet packet analysis with Mel frequency cepstrum analysis, and its robustness was improved. The experiment results show that the recognition rate based on wavelet packet analysis was improved by 6.78% compared with MFCC (DMel-frequency cepstrum coefficients) under noisy environment. When the SNR was 5dB, the recognition rate was up to 94.5%.