PROGRESS IN GEOGRAPHY ›› 2010, Vol. 29 ›› Issue (6): 747-756.doi: 10.11820/dlkxjz.2010.06.015

• Original Articles • Previous Articles     Next Articles

Mixed-pixel Decomposition and Super-resolution Reconstruction of RS Image

HU Maogui,WANG Jinfeng   

  1. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciencesand Natural Resources Research, CAS, Beijing 100101, China
  • Received:2009-06-01 Revised:2010-01-01 Online:2010-06-25 Published:2010-06-25


Remote sensing technology has been used in a wide range of applications, but the mixed-pixel phenomenon has been a persistent problem. In traditional classification, every pixel is considered a pure pixel and can be classified as only one type. This affects the accuracy and precision of results in applications. Recently, the problem has been studied by many researchers who have adopted many models and methods to decompose mixed-pixels and reconstruct super-resolution images from the low-resolution originals. In this article, we give a literature review of the development of mixed-pixel decomposition and super-resolution reconstruction. In accord with the main flow in the process, three aspects are reviewed: (1) endmember selection, (2) abundance estimation, and (3) super-resolution reconstruction. Endmember selection aims at selecting pure objects in the whole image range. Statistical methods and geometrical methods have been covered in detail for endmember selection. Abundance estimation of endmembers in pixels is another vital step attracting a great deal of research. It involves a number of new models and methods. We put an emphasis on variable endmember spectral mixture analysis and blind sources separation methods, which perform well and seem promising. Super-resolution reconstruction is based on the result of abundance estimation. How to maximize the spatial auto-correlation is the main objective when reconstructing super-resolution images. We review the most commonly used pixel-swapping method at length and discuss some problems presented in the study. Finally, some suggestions are brought forward for the mixed-pixel decomposition and super-resolution reconstruction of RS images.

Key words: abundance estimation, blind sources separation, endmember selection, mixed-pixel decomposition, super-resolution reconstruction