基于堆栈式稀疏自编码器的高光谱影像分类  

Hyperspectral Remote Sensing Image Chassification Using the Stacked Sparse Autoencoder

作　　者：戴晓爱[1] 郭守恒[1] 任淯 杨晓霞[1] 刘汉湖[1] 

机构地区：[1]成都理工大学地学空间信息技术国土资源部重点实验室,成都610059

出　　处：《电子科技大学学报》2016年第3期382-386,共5页Journal of University of Electronic Science and Technology of China

基　　金：国家自然科学基金(41201440,41071265,41102225);教育部高等学校博士学科点专项科研基金(20135122120009);四川省教育厅科研项目(15ZB0066);国土资源部地学空间信息技术重点实验室课题(KLGSIT2013-02)

摘　　要：为挖掘高光谱影像数据的内在光谱特征,该文基于深度学习理论,引用堆栈式稀疏自编码器构建原始数据的深层特征表达。首先通过稀疏自编码器,得到原始数据的稀疏特征表达。其次通过逐层学习稀疏自编码器构建深度神经网,输出原始数据的深度特征。最后将其连接到支持向量机分类器,完成模型的精调。实验结果分析表明:基于堆栈式稀疏自编码器的最优分类模型,总体精度可达87.82%,优于实验中的其他方法,证明了深度学习方法在高光谱影像处理中具有良好的分类性能。To extract rich features of hyperspectral image, this study explores the deep features of the raw data by using a stacked sparse autoencoder in the deep learning theory. First we create a sparse expression of raw hyperspectral image using sparse autoencoder. Then a deep neural network generating the deep features of raw data is built through learning stacked sparse autoencoder layer by layer. In addition, the deep feature-related model parameters are precisely calibrated by the statistical learning algorithm of the support vector machine （SVM）. The performance of the experiment indicates that the overall accuracy of classification model based on stacked sparse autoencoder reaches 87.82%, superior to other experimental methods. From our experiments, it follows that the deep learning theory and stacked sparse autoencoder are of high potential in hyperspectral remote sensing image classification.

关 键 词：深度神经网 特征提取 高光谱影像分类 堆栈式稀疏自编码器 支持向量机 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]