本文關(guān)鍵詞： 高光譜遙感 圖像分類 最優(yōu)代表向量 密度峰值聚類 冰川分類　出處：《北京理工大學(xué)學(xué)報》2017年10期 　論文類型：期刊論文

【摘要】：針對同物異譜現(xiàn)象以及分類過程中樣本代表性差、人工參數(shù)設(shè)置等原因?qū)е赂吖庾V遙感影像分類精度差的問題,提出了一種樣本集優(yōu)化的最優(yōu)代表向量分類法,對感興趣區(qū)中的樣本進(jìn)行密度峰值聚類提純,并對每類地物提純后樣本的均值向量集進(jìn)行隸屬度聚類擇優(yōu),獲取最優(yōu)代表向量集作為該類地物的中心向量,最終依據(jù)距離準(zhǔn)則進(jìn)行分類.通過對比實驗驗證,本文算法總體分類精度高于90%,表明最優(yōu)代表向量分類法能夠有效消除樣本差異性的影響,提高冰川分類精度.
[Abstract]:Aiming at the problem that the classification accuracy of hyperspectral remote sensing images is poor due to the heterospectral phenomenon of the same objects and the poor representation of samples and the setting of artificial parameters in the classification process, an optimal representative vector classification method for the optimization of sample sets is proposed. The samples in the region of interest are purified by peak density clustering, and the mean vector set of each kind of ground objects is selected by membership degree clustering, and the optimal representative vector set is obtained as the center vector of this kind of feature. Finally, the classification is based on the distance criterion. Through comparative experiments, the overall classification accuracy of this algorithm is higher than 90 points, which shows that the optimal representative vector classification method can effectively eliminate the impact of sample differences. Improve the accuracy of glacier classification.
【作者單位】： 北京科技大學(xué)自動化學(xué)院;
【基金】：高分辨率對地觀測系統(tǒng)重大專項基金資助項目 “十三五”武器裝備預(yù)研領(lǐng)域基金資助項目 中央高�；究蒲袠I(yè)務(wù)費專項資金資助項目(FRF-TP-15-117A1) 中國博士后科學(xué)基金資助項目(2016M600922)
【分類號】：TP751
【正文快照】： 冰川中蘊含著極其豐富的淡水資源,對全球氣候變化與生態(tài)平衡起著關(guān)鍵性的作用,但大部分冰川地處偏僻,面積寬廣,所以遙感技術(shù)被廣泛應(yīng)用于大尺度的冰川監(jiān)測中[1].近年來,全球流域冰川監(jiān)測研究主要采用Landsat ETM+、SPOT5、ASTER等多光譜遙感影像,而高光譜遙感影像應(yīng)用較少.然，

本文編號：1460305