摘要
提出了以多尺度连续小波变换值矩阵的奇异值作为识别特征矢量的方法,并利用该方法对湖底回波实测数据进行特征提取与分类。理论分析与仿真试验结果表明,相对于时间-小波能量和尺度-小波能量特征提取法而言,该方法可得到更好的类内紧致性和类间可分性,以及更佳的分类效果。
In our opinion, existing methods (scale-wavelet power spectrum and time-wavelet power spectrum) of extracting features of underwater echoes extract only a part of the feature information available in the coefficient matrix of CWT (Continuous Wavelet Transform). We present a method that we believe can exploit most of its feature information. Our better feature extraction method uses eigenvalues of the CWT coefficient matrix as feature vectors. All three methods construct feature vectors on CWT coefficient matrix, We explain our eigenvalue method in much detail in the full paper; we omit such explanation in this abstract. Features of data from lake experiments are extracted with these methods and sent to SVM (Support Vector Machine) classifier. We performed four experiments. In each experiment we had 144 samples. In these four experiments, the average numbers of samples correctly recognized and the percentages of correct recognition are : (1) eigenvalue method, 131 samples and 91.0 % ; (2) scale-wavelet power spectrum method, 113 samples and 78.3%; (3) time-wavelet spectrum method, 102 samples and 71. 0%.
出处
《西北工业大学学报》
EI
CAS
CSCD
北大核心
2006年第1期111-114,共4页
Journal of Northwestern Polytechnical University
关键词
连续小波变换
奇异值分解
尺度-小波能量谱
时间-小波能量谱
CWT (Continuous Wavelet Transform), eigenvalue, scale-wavelet power spectrum, timewavelet power spectrum