邢鈺佳 閆德勤 劉德山 王軍浩



摘? 要: 高光譜圖像的分類研究是高光譜圖像處理與應用的重要環節。為有效提取高光譜遙感圖像的空間信息和光譜信息,本文基于極限學習機提出新的研究。在模式識別和機器學習領域,極限學習機以其簡單、快捷和良好的泛化能力得到越來越多的關注。但由于在高光譜遙感圖像的學習過程中極限學習機缺乏對空間信息和光譜信息的有效提取,無法在分類中提供良好的分類結果。為此,基于譜局部信息的思想構造本文的研究框架,提出一種加權空-譜局部信息保持極限學習機分類算法。為驗證所提算法的有效性,本文在兩組常用的高光譜數據集Indian Pines和University of Pavia上進行實驗,通過與傳統的分類算法SVM和目前較為流行的分類算法KELM,KCRT-CK,MLR和LPKELM相比,本文算法具有較好的分類精度。
關鍵詞: 極限學習機;高光譜遙感圖像分類;加權空-譜;局部信息保持
中圖分類號: TP3 ???文獻標識碼: A??? DOI:10.3969/j.issn.1003-6970.2020.07.023
本文著錄格式:邢鈺佳,閆德勤,劉德山,等. 加權空-譜局部信息保持極限學習機的高光譜圖像分類算法[J]. 軟件,2020,41(07):113-119+135
Hyperspectral Image Classification Algorithm for Weighted Spatial SpectralLocality Information Preserving Extreme Learning Machine
XING Yu-jia, YAN De-qin, LIU De-shan, WANG Jun-hao
(Department of Computer Science, Liaoning Normal University, Dalian 116081, China)
【Abstract】: Classification of hyperspectral images is an important part of hyperspectral image processing and application. In order to effectively extract the spatial and spectral information of hyperspectral remote sensing images, this paper proposes a new study based on extreme learning machines. In the field of pattern recognition and machine learning, extreme learning machines have attracted more and more attention due to their simplicity, speed and good generalization capabilities. However, due to the lack of effective extraction of spatial information and spectral information during the learning process of hyperspectral remote sensing images, extreme learning machines cannot provide good classification results in classification. To this end, based on the idea of spectral local information, a research framework for this paper is constructed, and a weighted spatial spectral locality information preserving extreme learning machine classification algorithm is proposed. In order to verify the effectiveness of the proposed algorithm, this paper performs experiments on two commonly used hyperspectral data sets, Indian Pines and University of Pavia, and compares with the traditional classification algorithm SVM and the currently popular classification algorithms KELM, KCRT-CK, MLR Compared with LPKELM, our algorithm has better classification accuracy.
【Key words】: Extreme learning machine; Hyperspectral image classification; Weighted spatial spectral; Locality information preserving
0? 引言
高光譜遙感圖像(HSI)是由數百個光譜帶組成的3維立體圖像,如何從中提取大量信息應用于圖像分類是遙感圖像領域面臨的一項挑戰。目前,許多機器學習算法均被應用于高光譜遙感圖像的分類任務中,諸如K-近鄰(KNN)[1],Logistic回歸[2,3],人工神經網絡(ANN)[4]等。在這些方法中,支持向量機(SVM)[5]已經被證明在小樣本且包含噪聲的情況下具有最優的分類精度。然而在高光譜遙感圖像的分類任務中存在著較為明顯的Hughes現象[6],如果將原始頻段未經處理直接應用于機器學習算法,將會對分類精度產生較大的計算負擔。……