联合稀疏表示在水下目标分类中的应用  

Application of joint sparse representation in underwater target classification

作　　者：王佳维 许枫[1] 杨娟[1] WANG Jiawei;XU Feng;YANG Juan(Institute of Acoustics,Chinese Academy of Sciences,Beijing 100190;University of Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]中国科学院声学研究所,北京100190 [2]中国科学院大学,北京100049

出　　处：《声学学报》2022年第4期471-480,共10页Acta Acustica

基　　金：国家重点研发计划项目(2018YFC0824103);中国科学院战略性先导科技专项(XDA13030604);海南省重大科技计划项目(ZDKJ2020010)资助。

摘　　要：水下目标分类识别的性能受所选特征的限制,多特征往往可以获得更加稳定的结果,针对这一问题,提出了一种基于联合稀疏表示模型的水下目标分类识别方法。首先对水下目标回波信号提取3种具有信息互补性与关联性的特征:中心矩特征、小波包能量谱特征、梅尔频率倒谱系数特征,然后应用加速近端梯度法对联合稀疏表示模型进行优化,求解得到最优联合稀疏系数,最后根据最小误差准则确定目标类别。在消声水池开展模拟实验,对6类目标进行分类识别,结果表明:与传统算法相比,提出的算法具有更高识别准确率,并且其执行效率较传统算法有很大提升。The performance of underwater target classification and recognition is limited by the selected features,and the application of multi features is usually very helpful for the stability of classification results.To solve this problem,this paper proposes a method of underwater target classification and recognition via joint sparse representation model.Firstly,three kinds of features with information complementarity and correlation are extracted from underwater target echo signals:the central moments feature,the wavelet packet component energy feature and the Mel Frequency Cepstrum Coefficients feature,and then the joint sparse representation model is optimized by using the accelerated proximal gradient method,and the optimal joint sparsity coefficient is obtained,finally the class labels for test samples are determined via the minimum reconstruction error criteria.The simulation experiment is conducted in anechoic tank to classify to identify six kinds of targets.The results show that compared with the traditional algorithm,the proposed algorithm has higher recognition accuracy,and its execution efficiency is greatly improved.

关 键 词：梅尔频率倒谱系数 小波包能量谱 联合稀疏表示 分类识别 水下目标分类 信号提取 模拟实验 消声水池 

分 类 号：TB56[交通运输工程—水声工程]