基于改进的ISOMAP降维和改进的SAM高光谱图像分类
Hyperspectral Image Classification Based on Improved ISOMAP Dimensionality Reduction and Improved SAM

随着模式识别、机器学习、遥感技术等相关学科领域的发展，高光谱遥感影像分类研究取得了快速进展。高光谱影像具有高维度和高相关性，导致“维度灾难”。而高光谱降维的手段主要分为波段选择和波段提取。流形学习作为一种波段提取方法，在高光谱数据降维和分类中已被广泛使用。本文提出一种改进的ISOMAP降维和改进的SAM，并用Salinas-A数据集对比SAM直接分类和PCA降维后SAM分类(PCA-SAM)。比较了分类后整体和每个类别的精确度，发现新提出的算法在整体精确度都要高于SAM和PCA-SAM，并且发现在个别类别的精确度上有很大的提高。
With the development of related disciplines such as pattern recognition, machine learning, and remote sensing technology, research on the classification of hyperspectral remote sensing images has achieved rapid progress. Hyperspectral images have high dimensionality and high correlation, resulting in the “curse of dimensionality”. The methods for hyperspectral dimensionality reduction are mainly divided into band selection and band extraction. Manifold learning, as a feature extraction method, has been widely used in hyperspectral data dimensionality reduction and classification. This paper proposes an improved ISOMAP dimensionality reduction and an improved SAM, and compares the direct classification of SAM with the classification of SAM after PCA dimensionality reduction (PCA-SAM) using the Salinas-A dataset. The overall accuracy and the accuracy of each category after classification are compared. It is found that the newly proposed algorithm has higher overall accuracy than both SAM and PCA-SAM, and significant improvements in accuracy are observed in individual categories.