基于稀疏非负矩阵分解的低空声目标识别  

Low altitude acoustic target recognition based on sparse non-negative matrix factorization

作　　者：杨博[1] 杨立学 王志峰[1] 周印龙[1] YANG Bo;YANG Lixue;WANG Zhifeng;ZHOU Yinlong(The Third Research Institute of China Electronic Technology Group Corporation,Beijing 100015,China)

机构地区：[1]中国电子科技集团公司第三研究所,北京100015

出　　处：《声学技术》2020年第1期93-97,共5页Technical Acoustics

摘　　要：借鉴人耳听觉原理和特征学习的优势,提出了梅尔(Mel)频率谱提取和稀疏非负矩阵分解相结合的方法用于低空飞行目标声信号识别。首先,以不同目标的Mel频率谱为特征矩阵,利用稀疏非负矩阵分解方法学习得到各自的模板矩阵;然后,利用按列合并后的模板矩阵对训练/测试样本进行特征分解获得编码系数,该系数可作为分类特征;最后,结合不同目标的特点,采用分频段特征提取和顺序二类分类的方法进行多目标分类,并与Mel频率倒谱系数进行性能比较。结果显示,无论在单类目标辨识还是在多类目标分类中,稀疏非负矩阵分解方法均取得了更好的效果。A passive acoustical recognition technique of low altitude targets is studied in this paper.Based on the human auditory principles and the superiorities of feature learning,this technique combines the methods of Mel-scale spectrum extraction and sparse nonnegative matrix factorization.Firstly,the Mel-scale spectrums of different kinds of targets are taken as the feature matrices,their respective template matrices are obtained via learning with the sparse NMF technique.Then,the grouped template matrices are used for decomposing the signal features of each set of training/testing data to obtain coding coefficients,which are regarded as the feature vectors in classification model.Finally,according to the characteristics of different targets,the multi-band feature extraction and sequential binary classification procedures are used for multi-target classification,and the classical MFCC features are treated as the baseline to compare with the results of the sparse NMF method.The classification experiments demonstrate that the sparse NMF method achieves significantly better results in both single-class target identification and multi-class target classification.

关 键 词：低空声目标识别 梅尔频率倒谱系数 非负矩阵分解 

分 类 号：TN911.7[电子电信—通信与信息系统]