基于稀疏感知学习的高光谱遥感影像分类

项目名称： 基于稀疏感知学习的高光谱遥感影像分类

项目编号： No.61272282

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 张向荣

作者单位： 西安电子科技大学

项目金额： 62万元

中文摘要： 由于高光谱遥感影像具有地物信息复杂、大量有标记样本难以获取、波段多、数据量大等特性，从而导致其地物分类存在精度低，区域一致性差等应用瓶颈问题，对此，本项目充分挖掘高光谱遥感影像的图像空间、光谱空间的稀疏性，设计空-谱联合稀疏感知的高光谱图像分类；将稀疏感知和半监督学习相结合，提出压缩感知半监督学习框架，设计不平衡自适应稀疏度压缩学习分类器,半监督自适应字典的压缩学习分类器,经验映射稀疏表示分类器,和观测矩阵优化的维数约减算法，建立高光谱影像鲁棒、准确的分类方法。期望在稀疏感知学习分类器设计方法上有所突破，在高光谱遥感影像分类应用上取得实质性进展。

中文关键词： 高光谱遥感影像分类；稀疏表示；半监督学习；空谱特征学习；

英文摘要： Hyperspectral remote sensing image is characterized as complex materials, limited labeled pixels, large number of bands, and huge size of data, which result in low accuracy and bad uniformity in homegeneous regions in the hyperspectral image classification application. Exploiting the sparsity of spatial and spectral information in hyperspectral images, a joint spatial-spectral sparse sensing for classification is proposed. Also, by combining the sparse sensing and semisupervised learning, a compressive semi-supervised learning framework is presented. Specially, we will study the sparse coding classifier with unbalanced adaptive sparsity, task-driven semi-supervised dictionary learning based classifier, empirical kernel sparse coding classifier, and measurement matrix optimization based dimension reduction to get more robust and accurate hyperspectral image classification. By this study it is expected to get some innovations in the sparse sensing and learning based classification, and obtain substantial progress in hyperspectral remote sensing image classification.

英文关键词： Hyperspectral Remote Sensing Image Classification；Sparse Representation；Semi-supervised Learning；Spatial-spectral Feature Learning；