Research On Hyperspectrum Dimension Reduction Algorithm Based On Tdpca And Spiht Compression

The huge data of hyperspectral images produces many problems on data store, data transmission and data processing. Therefore, compression technology research has become hot today. The methods of data compression and dimension reduction of hyperspectral images are different from the normal images because of their own inter dependencies and spatial correlation. As for the dimension reduction and compression of hyperspectral images, this paper discusses a series of traditional and modern methods and then introduces a new dimension reduction method which aims at analyzing hyperspectral images based on Two Dimensional Principal Analysis. Since compression is aimed at advancing the compression ratio of images, according to the different interdependencies between wave bands, this paper introduces a compression method, which combines wave band arrangement, optimal linear predictor which makes the square error minimal and SPIHT, to rearrange the forecasting sequence before forecasting code. This method contains the following several aspects:Firstly, this paper introduces the imaging spectrometer and the concept of remote sensing imaging as well as the current domestic and foreign development situation of images compression technology and many criterions for the quality evaluation of images compression. In order to further improve the aims and effects of compression and dimension reduction, this paper chooses PSNR and MSE to evaluate the compression results. In addition, this paper introduces the three main methods for compressing the hyperspectral images and finds out their characters through comparing hyperspectral images and normal images. Also this paper introduces some traditional dimension reduction methods, including PCA etc. and a new TDPCA and RTPCA method based on these traditional methods. The new method collects the characteristics of hyperspectral images through multivariable linear transformation and reduces the dimension by applying Two Dimensional Principal Analysis. The experimental results show that the method achieved little computation, lower variance, greatly improved PSNR and classification accuracy with slight decline of MSE.At last, this paper introduces wavelet transform and its tree structure as well as the encoding methods including EZW and SPIHT which are advanced by analyzing the tree system of data after the change of the small wave of images. This paper provides the introduction and analysis of SPIHT in details and then introduces a new method which combines wave band arrangement, optimal linear predictor which makes the square error minimal and SPIHT to compress the hyperspectral images. This method is able to achieve enough PSNR and satisfactory MSE and calculation time. The reduced picture shows that this method can preserve maximal information of the original images.