HHT与神经网络在舰船目标特征提取中的应用

目标识别一直是水声领域的关键技术之一。将高阶累积量用于希尔伯特变换特征提取中,通过对舰船目标辐射噪声信号进行采集,得到舰船目标噪声信号,进而提取目标辐射信号各阶模态的相邻平均瞬时频率比、相对标准差、中心频率、平均强度、高阶矩和高阶累积量等作为特征,最终利用BP神经网络来实现对两类舰船目标的分类识别。通过对实际舰船目标噪声进行识别,验证了该舰船目标识别系统具有较好的识别效果。

The application of HHT and neural network in feature extraction of ship targets

Target recognition is one of the key techniques in underwater acoustic area. This article uses high-order cumulant and Hilbert transform for feature extraction, firstly gets the ship radiated noise from target ships, and then extracts the ratio of average instantaneous frequency between neighboring IMFs, relative standard deviation, center frequency, average intensity, high-order moment and high-order cumulant of different orders of IMFn （n=l-8）, finally recognizes and classifies two types of ship targets through BP neural network. Good recognition effect of this method has been verified through the classification tests for the actual ship radiated noise.