摘要
基于多波束海底声图像中多种特征信息数据的不同特点,以经典的基本统计算法、基于灰度共生矩阵的纹理分析以及基于功率谱比的Pace谱特征提取方法得到3组特征向量,并组合形成4个合成核以代替传统的单核形式,进而采用支持向量机(support vector machine,SVM)进行底质分类研究.通过海试数据处理对该方法进行评价和验证,结果表明该方法可获得比传统单核SVM更高的分类精度,具备实际应用前景.
A variety of feature data obtained by multibeam seabed acoustic image have different characteristics.In this paper,three groups of feature vectors have been obtained by three kinds of methods,include basic statistical algorithms,texture analysis based on grey level co-occurrence matrices（GLCM）and Pace spectral features extraction method,and have been combined four kinds of composite kernel forms instead of single kernel.Then,support vector machine（SVM）was used to achieve seabed sediment classification. The effectiveness of the method is discussed and validated by processing the sea experiment data.The results show that seabed classification using composite kernel SVM can achieve higher classification accuracy than traditional single kernel SVM method,and have the potential of practical application.
出处
《地球物理学进展》
CSCD
北大核心
2014年第5期2437-2442,共6页
Progress in Geophysics
基金
国家自然科学基金科学仪器基础研究专项(41327004)
青年科学基金项目(41306038)
水声重点实验室基金项目(9140C200105120C2001)
中央高校基本科研业务费专项基金项目(HEUCF140502)联合资助
关键词
合成核
支持向量机
声图像
底质分类
特征提取
composite kernel
support vector machine
acoustic image
seabed classification
feature extraction
