论文部分内容阅读
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。
In order to solve the problem of decreasing the recognition rate of target acoustic signal classification in water with engineering background, a modification method of Gammatone filter bank for peripheral model is proposed to solve the feature of auditory peripheral model. The narrowband noise characteristics obtained can obviously improve the performance of underwater target recognition. Firstly, the reasons for the decrease of recognition rate were analyzed. It was found that multi-channel data acquisition in acoustic engineering application resulted in the narrowing of signal frequency range, which caused the change of time-frequency characteristics of acoustic signal. Secondly, according to the auditory model, the frequency decomposition of the human ear basement membrane was simulated by the Gammatone filter bank. The low frequency information included important categories of target noise signals in water. The original auditory model features were interpolated. The number of channels and the center Frequency adaptive correction to get the target noise in a narrow band of 27-dimensional features, the modified model to more accurately reflect the target time-frequency characteristics. Finally, we use neural network classifier to experiment. The results show that the modified auditory model preserves the main information of the original relatively wide frequency band, and further improves the classification ability of the actual target. The recognition rate increases from 82.59% to 88.80%. In this paper, we design the design of Gammatone filter bank based on the effective receiving frequency band of the application platform to optimize the auditory perimeter model, analyze the multi-channel data using array element level, and focus on engineering applications. In the multi-channel data acquisition, Resulting in a decrease of the characteristic information amount of the signal, thereby causing the problem of degrading the acoustic feature recognition performance and the modified auditory model feature, thereby effectively improving the recognition effect of the target radiation noise in the water. This paper is of some reference value for researchers engaged in time-frequency analysis of passive sonar target recognition, active sonar target recognition and bandwidth limited multi-channel acoustic data acquisition.