分段线性光谱分类器的优化算法设计  

Optimizing Algorithm Design of Piecewise Linear Classifier for Spectra

作　　者：兰天鸽[1] 方勇华[1] 熊伟[1] 孔超[1] 李大成[1] 董大明[1] 

机构地区：[1]中国科学院安徽光学精密机械研究所遥感研究室,安徽合肥230031

出　　处：《光谱学与光谱分析》2008年第11期2726-2729,共4页Spectroscopy and Spectral Analysis

基　　金：国家自然科学基金项目(60678056)资助

摘　　要：快速准确地识别污染气体种类是光谱法环境监测技术对分类器的基本要求。分段线性分类器简单、计算量小,可以较好的逼近非线性分界面。文章根据最大化分类间隔的思想,结合分段线性分类器和线性支持向量机,设计了单边分段线性分类器优化算法。对某气体模拟剂光谱的分类实验表明,经过优化算法训练的分段线性分类器可以用较少的超平面逼近非线性分界面,而且得到更高的识别准确率。Being able to identify pollutant gases quickly and accurately is a basic request of spectroscopic technique for envirment monitoring for spectral classifier.Piecewise linear classifier is simple needs less computational time and approachs nonlinear boundary beautifully.Combining piecewise linear classifier and linear support vector machine which is based on the principle of maximizing margin,an optimizing algorithm for single side piecewise linear classifier was devised.Experimental results indicate that the piecewise linear classifier trained by the optimizing algorithm proposed in this paper can approach nonolinear boundary with fewer superplanes and has higher veracity for classification and recognition.

关 键 词：分段线性 光谱分类器 优化算法 线性支持向量机 

分 类 号：O433[机械工程—光学工程]