水中目标窄带噪声识别的听觉外周模型

为解决听觉外周模型特征在具有工程背景的水中目标声信号分类研究中识别率下降问题,提出了一种外周模型Gammatone滤波器组修正方法,获得的窄带噪声特征可明显提高水中目标识别性能。首先,分析了识别率下降原因,发现声学工程应用中多通道数据采集,导致信号频率范围变窄,而引起声信号的时频特征发生变化。其次,根据听觉模型用Gammatone滤波器组模拟人耳基底膜频率分解特性、低频信息包含水中目标噪声信号的重要类别特征,对原有的听觉模型特征进行插值,对滤波器组的通道数与中心频率进行适应性修正,得到目标噪声在较窄频带的27维特征,修正后的模型能够更精细地反映出目标时频特性。最后,采用神经网络分类器进行实验。结果表明,修正后的听觉模型保留了原较宽频带特征的主要信息,而且进一步提高了对实际目标的分类能力,识别率由原来的82.59%提高到88.80%。本文提出根据工程应用平台的有效接收频带优化听觉外周模型Gammatone滤波器组的设计,采用阵元级的多通道数据进行分析,侧重于工程应用,解决了多通道数据采集中,由于频带变窄,导致信号的特征信息量下降,进而引起声特征识别性能下降的问题,修正后的听觉模型特征,有效地提高水中目标辐射噪声的识别效果。本文对从事无源声呐目标识别、有源声呐目标识别、带宽受限的多通道声数据采集的时频特性分析研究人员具有一定的参考价值。 

An auditory periphery model for underwater target narrow-band noise recognition

In order to solve the problem of recognition rate decreasing when apply the auditory periphery feature to identify the underwater target with practical application background, a Gammatone filter bank adjustment method is proposed. The narrow-band noise feature obtained may improve the recognition result significantly. Firstly, the reason of recognition result decreasing is analyzed, which is due to the mechanism of multichannel data acquisition in acoustic engineering. The decrease of sampling rate may cause the narrow down of signal frequency range, which leads to time- frequency feature distortion. Then, the Gammatone filter bank is implemented to simulate the frequency decomposition characteristics of human ear basement membrane. Since the class information of the underwater target noise signal is mostly contained in the low frequency part, the auditory feature of the conventional model is interpolated and the channel number of the filter bank and the central frequency of each frequency band are adjusted accordingly to obtain a 27-dimensional feature vector of the narrow-band target signal. The adjusted model may reflect the target's time-frequency feature more precisely. Finally, the performance of the auditory feature is tested by Neural Network classifier. The experiment result shows that the modified auditory model is more effective than the conventional ones. The majorinformation contained in broadband signals is reserved and the classification ability for real targets is further enhanced. The recognition result has increased from 82.59% to 88.80%. The proposed Gammatone filter bank adjustment method is based on effective frequency band of the data by practical application platforms. The multichannel data from the array element is used for data analyzing and the problem of recognition result decreasing due to narrow-band feature distortion is solved. The modified auditory feature may effectively improve the recognition rate for underwater target radiated noise signals. This article has a certain reference value for people doing studies and time-frequency feature analysis for acoustic data acquired by multiple active/passive sonar target recognition channels with limited frequency band.