Research On Some Key Technologies Of Hyperspectral Image Processing

Hyperspectral image provides abundant spatial and spectral information about the observed object,and it has been widely used in military and civil fields. In order to make full use of its advantage,there are important theoretical meanings and application value to develop efficient analysis andprocessing methods. And the methods are suitable for the characteristics of hyperspectral image. Onthe basis of previous research results, research on hyperspectral image compression, denoising, fusion,target detection and classification techniques has been done in this paper, and described as follows:Firstly, a compression method based on three-dimensional integer wavelet transform and waveletsupport vector regression is discussed. After the hyperspectral image is decomposed bythree-dimensional integer wavelet transform, the wavelet support vector regression can learn from thedependency between the coefficients of high frequency subbands. And the wavelet coefficients arerepresented sparsely by support vectors. As a result, the coefficients of high frequency subbands arecompressed. The wavelet coefficients are effectively treated in this method. It doesn’t only obtain lowbit rate, but also guarantees the quality of the compressed image.Then a hyperspectral image denoising method based on nonsubsampled contourlet transform(NSCT) and kernel principal component analysis (KPCA) is given. Hyperspectral image of each bandis decomposed by NSCT, and the NSCT coefficients are processed by KPCA. The proper principalcomponents are selected for KPCA reconstruction according to noise features. Experimental resultsshow that the proposed method can suppress noise interference in hyperspectral image, and preservethe useful information of original data more completely.And then, a fusion method for multispectral and panchromatic images using NSCT and regionsegmentation is studied. In this method, the NSCT coefficients of high and low frequency subbandsare fused by different rules, respectively. For the high frequency subbands, the fusion rules are alsounalike in the smooth and edge regions. The two regions are segregated in the panchromatic image,and the segmentation is based on particle swarm optimization (PSO). The experimental results areevaluated by both subjective and objective criteria. It is shown that the proposed method can obtainsuperior results to others.Next, a target detection method using projection pursuit based on chaotic PSO is introduced.Skewness and kurtosis which are susceptible to outliers are chosen to design the projection index. Andchaotic PSO is applied to find optimal projection direction. Chaotic PSO can speed up the process ofprojection pursuit and gets more accurate optimal projection direction. A large number of experimental results show that the proposed method not only detects target in hyperspectral imagesmore effectively, but also significantly reduces the running time.Finally, a hyperspectral image classification method based on empirical mode decomposition(EMD) and relevance vector machine (RVM) is presented. In order to solve the hyperspectral image’sredundant problem, adaptive band selection method is used to realize the process of hyperspectraldata’s redundant information, and EMD is used for feature extraction. Then the processedhyperspectral data is classified by RVM. The proposed method not only improves the classificationaccuracy of hyperspectral image, but also reduces the number of support vector and accelerates thehyperspectral image classification.