一种基于超限稀疏多项逻辑回归和奇异谱分析的高光谱遥感影像分类方法

Classification method of hyperspectral remote sensing image based on extreme sparse multinomial logistic regression and singular spectrum analysis

由于高光谱图像存在大量噪声,超限稀疏多项逻辑回归无法分析高光谱图像的内在结构,其适用性有待进一步提高,为解决超限稀疏多项逻辑回归不能有效应对噪声的问题,提出了一种基于超限稀疏多项逻辑回归和奇异谱分析的高光谱遥感影像分类方法:首先对高光谱遥感影像数据集进行归一化处理以消除数据量纲的影响,随后利用奇异谱分析对影像进行有效信息提取及噪声剔除,最后通过超限稀疏多项式逻辑回归对处理过的数据实现分类。采用多种不同数量的训练样本进行实验,并与3种常用分类算法进行对比分析,评价了本文方法的有效性和鲁棒性。结果显示,本文方法在各类训练样本情况下相比于其他分类方法,其总体分类精度皆有一定程度的提升。

Extreme Sparse Multinomial Logistic Regression can not analyze the internal structure of hyperspectral image as noises in hyperspectral images refrain the performance of ESMLR. When ESMLR,which is sensitive to noise,is improved, the performance of ESMLR is better. A hyperspectral remote sensing image classification method based on Extreme Sparse Multinomial Logistic Regression(ESMLR) and Singular Spectrum Analysis(SSA) is proposed in this paper. Firstly,the hyperspectral remote sensing image dataset is normalized to eliminate the influence of the dimension of data. Then,singular spectrum analysis performs effective information extractions and thus removes the noise in hyperspectral images. Finally,the processed data is classified by extreme sparse multinomial logistic regression. In order to evaluate the effectiveness and robustness of the proposed method,a variety of different numbers of training samples are used to carry out experiments,and compared with three commonly used classification algorithms. The experiment results show that the overall classification accuracy of the proposed method is improved to a certain extent compared with other classification algorithms under various training samples.