摘要
模糊支持向量机是模糊技术与支持向量机的有机结合,其关键步骤是确定模糊隶属度函数。现有方法大多利用距离这一相似性测度从不同角度构造隶属度函数,实现过程比较复杂。对于高光谱数据的光谱特性,用距离表征地物的光谱亮度差异较为合适,但天气、光照强度等因素对这种亮度影响很大。相比之下光谱间的角度受亮度的影响很小,作为相似性测度更为可靠。针对这种地物光谱角度特性,在模糊最小二乘支持向量机(FLS-SVM)中,用核光谱角余弦作为相似性测度来构造模糊隶属度函数,仿真结果表明能够有效地提高最小二乘支持向量机(LS-SVM)高光谱图像分类性能。
Fuzzy support vector machine （FSVM） is a combination of fuzzy technology and support vector machine. A key for it is how to set the fuzzy membership. Presently, most methods use the distance as a similarity measure, in all dimensions, to derive the fuzzy memberships function together with a complex procedure. Distance measure is fit for representing spectral luminance difference between different types of terrain in consideration of the hyperspectral date. While, at the same time, there always exits some other factors like the weather and light intensity and so on, having a worse effect on luminance. It is a more reli- able Similarity measure for spectral angle, when compared to the distance measure. In this paper, we focus on using the kernel spectral angel cosine as a similarity measure to realize FLS-SVM. It＇ s easy to process, the simulation results indicate it effectively raise the performance of LS-SVM.
出处
《黑龙江大学工程学报》
2011年第4期78-83,共6页
Journal of Engineering of Heilongjiang University
基金
国家自然科学基金项目(61077079
60802059)
高等学校博士学科点专项基金(20102304110013)
哈尔滨市优秀学科带头人基金(2009RFXXG034)
关键词
高光谱图像分类
SVM
相似性测度
核光谱角余弦
模糊隶属度
hyperspetral image classification
SVM
similarity measure
kernel spectral angel cosine
function of fuzzy membership
