首页 >  , Vol. , Issue () : -

摘要

全文摘要次数: 206 全文下载次数: 297
引用本文:

DOI:

10.11834/jrs.20210354

收稿日期:

2020-08-17

修改日期:

2021-06-27

PDF Free   EndNote   BibTeX
高光谱遥感影像降维:进展、挑战与展望
苏红军
河海大学
摘要:

高光谱遥感影像数据具有高维特征、信息冗余、不确定性显著、小样本、空谱合一等特征,对其进行数据处理面临巨大挑战,高光谱遥感影像降维是高光谱遥感的重要研究方向之一。本文对当前高光谱遥感影像降维的相关研究进展进行了综述,在介绍高光谱遥感数据特点的基础上,重点从特征提取和特征选择两方面对高光谱遥感影像降维的最新研究和前沿进展进行了系统性综述;并从特征可分性、特征质量评价、特征数目确定、多特征优化以及需求驱动的特征选择等方面分析了高光谱遥感影像降维面临的挑战。随着智能化高光谱遥感的发展,高光谱遥感影像智能降维成为未来的发展方向,同时其发展将兼顾多特征质量评估与优选、搜索策略优化、满足应用需求等多目标的需求;随着高光谱遥感数据获取能力的提升和深入应用,高光谱遥感影像降维将会发挥重要而不可替代的作用。

Dimensionality Reduction for Hyperspectral Remote Sensing: Advances, Challenges and Prospects
Abstract:

Hyperspectral imaging can provide narrow bands and continuous spectrum information; however, hyper-spectral image data have the characteristics of high dimensionality, rich features, information redundancy, small samples, significant uncertainty and so on, which results difficulties for hyperspectral image data pro-cessing. Dimensionality reduction of hyperspectral remote sensing is one of the important topics for hyper-spectral image data processing. Hyperspectral image data have hundreds of bands and can provide rich in-formation, however, there is a strong correlation between different bands, so that result in data redundancy. Therefore, during the processing of hyperspectral data, there will be curse of dimensionality problem, such as increasing of time complexity and overfitting of the prediction model due to the increase of the spectral fea-ture dimension. For the most important, the number of training samples available for hyperspectral remote sensing images is small, and the feature dimension is much larger than the training sample. With the increase of feature dimensionality, the classification accuracy will increase first and then decrease, that is, “Hughes” phenomenon. Therefore, how to make full use of the rich information of hyperspectral images data and solve the problem of high feature dimension through certain methods has become one of the key issues in the research of hyperspectral imaging data processing. The dimensionality reduction of hyperspectral remote sensing image is a technology to reduce the dimensionality of hyperspectral imaging by using feature extrac-tion or band selection while retaining effective information or features as much as possible. Feature extrac-tion methods use projection transformation method to map hyperspectral data from high-dimensional space to low-dimensional space, such as principal component analysis, linear discriminant analysis, independent component analysis, manifold learning and deep learning-based methods. Feature selection is to eliminate redundant bands without changing the original feature structure, and find representative feature band sub-sets, such as selection based on information measurement and feature correlation. With the development of new technologies, evolutionary algorithms and intelligent algorithms such as genetic algorithm, ant colony algorithm and firefly algorithm have also been applied for hyperspectral remote sensing dimensionality re-duction. This article systematically summarizes and reviews the current advances of dimensionality reduction for hyperspectral remote sensing, especially for feature extraction and feature selection. For feature extraction, we reviewed the advances of index & parameters-based, projection & transformation-based, band combina-tion-based, spatial algorithm-based, manifold learning-based and deep learning-based feature extraction al-gorithms. For bans selection, the advances for information measurement, search strategy, optimized band number, multi-feature quality assessment and optimization algorithms are reviewed. The challenges of di-mensionality reduction for hyperspectral remote sensing are analyzed from five aspects: feature separability, feature quality evaluation, feature number determination, multi-feature optimization and problem-oriented feature selection. With the development of intelligent hyperspectral remote sensing, intelligent dimensionality reduction will be one of the hottest topics. At the same time, multi-feature quality assessment, search strate-gy optimization, application requirements will attract special attentions in the future. Dimensionality reduc-tion of hyperspectral remote sensing will play an important and irreplaceable role in hyperspectral image da-ta acquisition and applications.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮