计算机仿真
計算機倣真
계산궤방진
COMPUTER SIMULATION
2009年
12期
164-167
,共4页
高光谱遥感%图像预处理%支撑向量机%核函数
高光譜遙感%圖像預處理%支撐嚮量機%覈函數
고광보요감%도상예처리%지탱향량궤%핵함수
Hyper- spectral remote sensing%Image preprocessing%Support vector machines (SVM)%Kernel function
高光谱遥感图像具有维数高的特点,当样本较少时,利用传统的统计识别方法分类,分类精度低.可支撑向量机(SVM)能解决小样本、高维、非线性分类问题.采用归一化法对原始图像做预处理,再分析不同的SVM核函数对分类精度的影响;并把SVM与最小距离法,马氏距离法等的分类结果进行比较.结果表明SVM的核函数类型对分类正确率影响不大,其分类精度高于传统的统计识别方法.
高光譜遙感圖像具有維數高的特點,噹樣本較少時,利用傳統的統計識彆方法分類,分類精度低.可支撐嚮量機(SVM)能解決小樣本、高維、非線性分類問題.採用歸一化法對原始圖像做預處理,再分析不同的SVM覈函數對分類精度的影響;併把SVM與最小距離法,馬氏距離法等的分類結果進行比較.結果錶明SVM的覈函數類型對分類正確率影響不大,其分類精度高于傳統的統計識彆方法.
고광보요감도상구유유수고적특점,당양본교소시,이용전통적통계식별방법분류,분류정도저.가지탱향량궤(SVM)능해결소양본、고유、비선성분류문제.채용귀일화법대원시도상주예처리,재분석불동적SVM핵함수대분류정도적영향;병파SVM여최소거리법,마씨거리법등적분류결과진행비교.결과표명SVM적핵함수류형대분류정학솔영향불대,기분류정도고우전통적통계식별방법.
The characteristic of hyper- spectral remote sensing data is high dimensional. The accuracy of tradi-tional classification methods is always unsatisfactory. Support vector machine (SVM) is a good method for solving limited sample, high dimensional, and nom- linear classification problem. The image data are preprocessed by Im-age normalization method, the influences of different kernel functions on classification accuracy are analyzed, and the classification results by SVM are compared with that of other methods (such as Minimum Distance method, Mahal-anobis Distance Method). Experiment results show that the classification accuracy is almost identical for different kernel functions, and VM method has better classification and recognition accuracy than traditional algorithms.
