超谱图象的压缩和分类系统及其相关技术的研究

According to the combination of image and spectral characteristics for hyperspcetral images, compression and classification and their relative technologies of hyperspectral images are researched in this project. Adaptive supspace decomposition and recursive subspace decomposition methods are proposed based on the fact that interband correlations of hyperspectral images tend to be higher than intraband correlations, which concentrate representative features of different ground materials in different subspaces.Thus the computing complexity is reduced and useful details are preserved. On the basis of subspace decomposition, the new methods of feature fusion classification based on wavelet multiresolution analysis and decision fusion classification based on consensus rule are proposed, which are suitable for hyperspectral image. At the same time, aiming at resolving huge volume problem of hyperspectral images, a fast compression method based on second generation wavelet transform is studied. Moreover, combining the spectral and spatial correlations of hyperspectral images, the RBP/JPEG (Recursive Bidirectional Prediction/JPEG) algorithm for hyperspectral image compression is proposed, which can insure the image reconstruction quality even in high compression ratio.This project has been overfulfilled, totally 16 papers were published on both national and international publications，3 relative papers were published during the application of this project. 6 of them were indexed by SCI, 12 by EI, and 4 by ISTP.Supporting by this project, the staff took active part in the international academic cooperation and intercommunication, and had attended several international conferences. Based on this project, 3 graduates (1 doctor and 2 masters) got their academic degrees..In order to meet the requirement of future war, the results of this project have been applied to the relative researching work of one national defense 863 project presently, solving the problem of interrelated background recognition..This project has been passed the appraisement organized by the Commission of Science Technology and Industry for National Defense. The appraisement conclusions are as follows: according to the plan, the staff have been overfulfilled the task, the theories are correct, the experiments are substantial, the data are complete, the methods are innovative. The production of this project has great value of generalization and application, which keeps ahead in our country and reaches international advanced level.

针对超谱图象的特点，该课题首先研究图象本身的特性，特别是超谱各波段之间的相关性；同时研究适合于超谱图象的量化方法及特征提取方法；并研究新的编码和分类方法以及超谱图象压缩和分类方法的评价准则。最终目标是实现一个技术先进、性能高、算法简单的超谱图象压缩及分类系统。该课题的研究对超谱图象的进一步应用将有重要的理论和实践意义。