高光谱遥感图像空谱联合分类方法研究

Spectral-spatial joint classification method of hyperspectral remote sensing image

在遥感影像研究领域里,高光谱数据分类是一个热点问题。近年来,在这个问题上涌现出很多研究方法,然而,大多数方法都是用浅层的方法提取原始数据的特征。将深度学习的方法引入高光谱图像分类中,提出一种新的基于深信度网络(DBN)的特征提取方法和图像分类架构用于高光谱数据分析。将谱域-空域特征提取和分类器相结合提高分类精度。使用高光谱数据进行实验,结果表明该分类器优于当前的一些先进的分类方法。此外,本文还揭示了深度学习系统在高光谱图像分类研究中具有的巨大潜力。

In remote sensing image research area, hyperspectral data classification is a hot topic. In recent years, many study methods for this issue emerge; however, the majority of the methods adopt the shallow layer method to extract the characteristics of original data. In this paper, the deep study method is introduced in the hyperspectral image classification; a new characteristic extraction method and image classification construction based on deep belief network （DBN） is proposed, and used in hyperspectral data analysis. The spectral-spatial feature extraction and classifier are combined together to achieve high classification accuracy. Experiment was carried out using the hyperspectral data; experiment results indicate that the proposed classifier is superior to some current advanced classification methods. In addition, this paper also reveals that the deep learning system has great potential in the study of hyperspectral image classification.