Research Of Satellite Cloud Image Fusion Based On Multiscale Geometric Analysis And Its Infulence To Typhoon Center Location

China is one of the countries which are seriously affected by tropical cyclones in the Northwest Pacific. The southeast coast is attacked by typhoon in every summer. And it has caused serious casualties and economic losses. Therefore, it is necessary for us to make a deep research to the typhoon. It is very important to analyze and monitor the important information during the process of the emergence, development and dissolution in order to reduce the loss. Nowadays, the satellite cloud image is the most important means to monitor the typhoon. The extraction, forecast analysis and computer implementation to the data of the meteorological satellite cloud image is the key technique of meteorological satellite cloud processing. In the process of the acquisition, processing and transmission, the satellite cloud image is often influenced by the interference of the imaging device and noise from external environment. It results in degrading in image quality. It is very important to reduce noise in the satellite cloud image for monitoring typhoon based on satellite cloud image. In addition, there are some multiple scanning radiometers which are corresponding to different channels in the meteorological satellite. After transformation, different channels will get satellite cloud images which contain different information. The different characteristics of typhoon cloud images are contained in different channel cloud images. If we can fuse the characteristic information of different channels, it will make use of the information in each channel to extract the fearture of the typhoon cloud images. This will provide a better basis in typhoon monitoring and prediction. This paper mainly focuses on the multiscale geometric analysis (Tetrolet transform and Shearlet transform) and their applications in satellite cloud image denoising and fusion. This paper includes three research works as follows:(1) An efficient satellite image denoising method which is based on Tetrolet transform and partial differential equations (PDE) has been proposed. The proposed method is degined by combining PDE with generalized cross validation (GCV) in Tetrolet transform domain. This will result in better denoising image quality. Tetrolet transform is a kind of adaptive multiscale geometric analysis tool, and it has good sparse approximation performance. It can keep the important edges and details in the cloud images well. In this algorithm, GCV theory is introduced to determine the optimal denoising threshold in the Tetrolet transform domain when it doesn’t know the variance of noise. PDE denoising model is used to smooth the block effect in the Tetrolet transform. It can keep the local features of an image well. The denoising image by the proposed algorithm is compared with the results of other five image donoising methods which are based on multiscale techniques. Experimental results show that the peak signal-to-noise ratio (PSNR) value of the proposed algorithm is the best, and the visual quality of denoised image is good and the detail is kept well. And this will result in good quality of image fusion.(2) An efficient multi-channel satellite cloud image fusion method in the Tetrolet transform has been proposed. Firstly, image histogram equalization is used for the multi-channel satellite cloud images to enhance their global contrast. Secondly, Tetrolet transform is implemented to source images and obtain the low frequency coefficients, high frequency coefficients and its corresponding covering distribution values of the image. Differenet fusion rules are desinged to low frequency and high frequency components. Laplacian pyramid algorithm is used to decompose the low frequency components in the Tetrolet domain. The mean value is used on its top layer, and the maximum absolute values on the other layers are used to fuse low frequency components. For high frequency components, the maximum standard deviation of the high frequency parts is used to fuse high frequency components in each block in Tetrolet domain. And the covering distribution values also consider their corresponding parts as the high frequency. Finally, inverse Tetrolet transform is used to obtain the final fusion image. The proposed image fusion method is compared with six similar image fusion methods:Laplacian pyramid, classical discrete orthogonal wavelet transform, Curvelet transform, NSCT (Non-Sampled Contourlet Transform) and Shearlet transform. The subjective and objective evaluations of the fusion images have been discussed. The fusion cloud images are used to determine the center position of the eye and non-eye typhoons and the location error values. The experimental results show that the overall performance of the proposed method is the best compared with other similiar methods. The center location accuracy by using the fusion cloud image by the proposed method can be improved.(3) A multi-channel satellite cloud image fusion method based on Shearlet transform has been proposed. Since the Shearlet transform can decompose images in any scale and any orientation and it has strong ability to approach the edge, Shearlet transform is introduced into the satellite cloud image fusion. For the low frequency components in Shearlet domain, Laplacian pyramid algorithm is used to decompose the subimages. The mean value is used on its top layer, and the maximum absolute values on the other layers are used to fuse low frequency components. For the high frequency components, a novel image fusion rule which combines the information entropy, average gradient and standard deviation of each high frequency subimage is designed. In order to enhance the details of images, a nonlinear enhancement operation is used to enhance the fusion high frequency subimages. The proposed image fusion algorithm is compared with five similar image fusion algorithms which are based on multiscale techniques. These fusion images are evaluated by the subjective and objective methods. A new evaluation index is proposed in order to evaluate the fusion cloud images. In order to verify the efficiency of the proposed algorithm, fusion images are used to determine the center position of the eye and non-eye typhoons. The experimental results show that the overall performance of the proprosed method is the best compared with other similiar methods. The center location accuracy by using the fusion cloud image by the proposed method can be improved.