基于空谱融合和随机多图的高光谱遥感图像农作物分类  

Crop Classification in Hyperspectral Remote Sensing Images Based on Spatial-spectral Fusion and Random Multi-graphs

作　　者：聂萍 李飞 杨昭 汪国强[1] NIE Ping;LI Fei;YANG Zhao;WANG Guoqiang(School of Electronic Engineering,Heilongjiang University,Harbin 150080,China)

机构地区：[1]黑龙江大学电子工程学院,黑龙江哈尔滨150080

出　　处：《无线电工程》2024年第5期1205-1216,共12页Radio Engineering

基　　金：国家自然科学基金(51607059);黑龙江省自然科学基金(QC2017059)。

摘　　要：针对高光谱遥感图像复杂农作物分类问题,提出了一种基于空谱融合和随机多图的高光谱遥感图像农作物分类方法。通过使用一种潜在特征融合和最优聚类(Latent Features Fusion and Optimal Clustering Framework,LFFOCF)的波段选择方法和分段主成分分析(Segmented Principal Component Analysis,SPCA)进行光谱降维,采用多尺度二维奇异谱分析(2-D-Singular Spectrum Analysis,2-D-SSA)应用于降维图像,以提取不同尺度的空间特征。将多尺度空间特征与主成分分析(Principal Component Analysis,PCA)得到的全局光谱特征融合送到随机多图(Random Multi-Graphs,RMG)中进行分类。在印度松树、萨利纳斯和龙口数据集上,所提出的方法与一些现有的方法进行了对比实验。实验结果表明,该方法的类别精度(Class Accuracy,CA)、总体分类精度(Overall Accuracy,OA)、平均分类精度(Average Accuracy,AA)和Kappa系数优于这些方法。To deal with the problem of complex crop classification in hyperspectral remote sensing images,a new crop classification method for hyperspectral remote sensing images is proposed based on spatial-spectral fusion and random multi-graphs.A band selection method of Latent Features Fusion and Optimal Clustering Framework(LFFOCF)and the Segmented Principal Component Analysis(SPCA)are used for spectral dimensionality reduction.Multiscale 2-D-Singular Spectrum Analysis(2-D-SSA)is applied to dimensionally reduced images to extract spatial features at different scales.Multiscale spatial features and global spectral features obtained by Principal Component Analysis(PCA)are fused to Random Multi-Graphs(RMG)for classification.The proposed method is compared with some existing methods on the datasets of Indian pine,Salinas and WHU-Hi-LongKou.Experimental results show that the proposed method outperforms these methods in terms of Class Accuracy(CA),Overall Accuracy(OA),Average Accuracy(AA)and Kappa coefficients.

关 键 词：高光谱遥感图像 农作物分类 空谱融合 随机多图 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]