最小二乘支持向量机的两点改进  

Two improvements for least squares support vector machines

作　　者：王立国[1] 赵亮[1] 石瑶[1] 路婷婷[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2015年第6期847-850,870,共5页Journal of Harbin Engineering University

基　　金：国家自然科学基金资助项目(61275010);国家教育部博士点基金资助项目(20132304110007);黑龙江省自然科学基金资助项目(F201409);中央高校基本科研业务费重大资助项目(HEUCFD1410)

摘　　要：最小二乘支持向量机以其较好的性能得到了广泛应用,但仍存在2点不足:一方面,最小二乘支持向量机将所有训练样本都作为支持向量参与未知样本的分类,导致该算法在泛化过程中处理速度较慢;另一方面,最小二乘支持向量机主要利用光谱数据进行训练和分类,忽略了对地物空间信息的挖掘,影响了分类精度。为此,提出一种基于库伦引力模型的样本缩减策略,在此基础上将分类结果与基于空间信息的分类器相融合,由此产生的新分类器可以有效解决以上两方面的问题。实验表明了新分类模型在分类精度与速度方面的优势。The least squares support vector machine（ LSSVM） has been widely used due to its better performance.However,LSSVM still has two deficiencies. On the one hand,it sacrifices the speed in the generalization process because it treats all the training samples as the support vector set to participate in classification of unknown samples.On the other hand,LSSVM mainly uses spectral data for training and classification,and it ignores mining on the spatial information of a terrestrial object,which influences classification precision. Therefore,a sample reduction strategy based on the Coulomb gravity model is proposed. And on this basis,a new classifier model which merges the above classification result with the classifier based on spatial information can solve the two problems mentioned above. Experimental results indicate that this new classifier performs better than the standard LSSVM both in classification accuracy and speed.

关 键 词：高光谱 支持向量机 样本缩减 马尔科夫随机场 空间信息 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]