基于鉴别流形正则化最小二乘分类的钴结壳超声识别

摘要

The surface of cobalt crust contains plenty of nonlinear manifold fearures. In order to solve ultrasonic recognition of cobalt crust using its manifold features,a lind of Discriminatively Regularized Least-Squares Classifier(DRLSC) is irtroduced in this paper. At first,the fact is found that the original DRLSC is not of convexity in general. And then,based on Ho-Kashyap leastsquares algorithn,a modified DRLSC which is of convexity and its kernelized edition(Kernel RRLSC,KDRLSC) are proposed.At last, the proposed classifier is used in ultrasonic recognition of cobalt crust. The experimental results show that the echo recognition correct rates of cobalt crust is improvedl using the proposed KDRLSC classifier.%为解决富含非线性流形特征的水下钴结壳超声识别问题,尝试引入一类鉴别流形正则化最小二乘机(Discriminatively Regularized Least-Squares Classifier,DRLSC).首先分析了原始DRLSC,指出其对应优化问题的非凸性.然后结合Ho-Kashyap分类器,提出了具有凸性的DRLSC模型,并给出了该模型的核化版本(Kernel DRLSC,KDRISC).最后,将该模型应用于水下钻结壳超声识别中.实验结果表明,采用本文的KDRLSC进一步提高了钴结壳的识别分类正确率.