舰船辐射噪声的响度和音色特征模型  

Loudness and timbre features model of ship radiated noise

作　　者：曹红丽[1] 方世良[1] 

机构地区：[1]东南大学水声信号处理教育部重点实验室,南京210096

出　　处：《东南大学学报（自然科学版）》2013年第2期241-246,共6页Journal of Southeast University：Natural Science Edition

基　　金：国家重大基础研究发展计划(973计划)资助项目(6131222);国家自然科学基金资助项目(11104029)

摘　　要：首先分析模拟舰船辐射噪声的归一化功率和Moore响度变化趋势,进一步用海试数据验证Moore响度的周期性变化特点,分析Moore响度周期性变化的原因,提取Moore响度均值作为一维特征.然后分析研究音色中有效的谱质心特征,根据谱质心的数学定义推导出谱质心带宽、谱质心斜度以及谱质心峰度的数学公式,举例分析其物理含义,并提出基于六分段频带的谱质心、谱质心带宽、谱质心斜度以及谱质心峰度共24维的音色特征模型.将Moore响度特征模型和音色特征模型相结合,从组合的25维特征的均值和方差分布可知特征具有一定的可区分性.最后设计3层BP神经网络分类器,海试数据表明基于Moore响度和音色的特征模型提取方法是有效的,且有利于提高分类识别率.First,the normalized power and Moore loudness change trend of the simulated ship radiated noise is analyzed.Furthermore,the Moore loudness cyclical change is tested by the sea trial data,and the Moore loudness mean is extracted as one dimensional characteristic.Then the effective timbre spectrum centroid characteristics are studied,and the formulae of spectral centroid bandwidth,spectral centroid skewness and spectral centroid kurtosis are deduced according to the mathematical definition of spectrum centroid.The physical meanings of spectrum centroid,spectral centroid bandwidth,spectral centroid skewness and spectral centroid kurtosis are illustrated by an example,and a 24-dimensional features model based on spectral centroid,spectral centroid bandwidth,spectral centroid skewness and spectral centroid kurtosis of six segmented frequency bands is proposed.The Moore loudness feature model is combined with the timbre feature model,and the combined characteristics have certain separability,which can be seen from the mean and variance distributions of 25-dimensional characteristics.Finally,a three-layer BP neural network classifier is designed,and the sea trial data simulation results indicate that the feature extraction method is effective and helpful to improve the recognition rate.

关 键 词：Moore响度 谱质心 谱质心带宽 谱质心斜度 谱质心峰度 BP神经网络 

分 类 号：TN911.7[电子电信—通信与信息系统]