Research Of Pixel-level Remote Sensing Image Fusion Method

Remote sensing is a technology that developed rapidly from the middle of last century. It has been widely used in environmental, atmospheric, resources, oceans, geology, agriculture, forestry and other fields because it has some observation features that range widely, speed quickly, period shortly, methods diversely and has clooected redundant information about the earth. Remote sensing images are the primary means to record information. For different reasons, human access to remote sensing images with different spectral bands, spatial resolution and time resolution, they are reflection of different characteristics of target observed. In practice, images with different spectral bands, spatial resolution and time obtained are often needed to be fused to produce more comprehensive, accurate information so as to provide a robust basis for decision-making. This dissertation mainly aims at the research of pixel level fusion methods of multi-spectral images and panchromatic image, visible images and synthetic aperture radar (Synthetic Aperture Radar, SAR) image, the main innovations of this dissertation are summarized as follows:(1) For the fusion of multi-spectral images and panchromatic image, IHS (Hue-Intensity-Saturation) transform fusion method is an important method and has been integrated into a number of remote sensing image processing softwares. Studies show that the fusion method based on Wavelet+IHS transform is the one of best methods, but the fusion result of IHS transform consist of serious spectral distortion. The researchers have studied this problem deeply and proposed many measures to improve. This paper analyzes the various IHS transform fusion method previously and considers the spectrum distortion is caused by the large difference between I component of IHS transformed and the high resolution panchromatic image. It is taken into account that the previous method does not consider the complexity of satellite imaging and includes many uncertainties, using the statistical method to determine the transformation I calculate the weight of components are proposed here. Fusion result evaluation shows the proposed method could preserve spectral character better than previous IHS transform in the case of maintaining ability of integrate spatial details of panchromatic image. But compared to ideal images, fusion result still has some loss.(2) During the fusing of multi-spectral images and panchromatic image, it is thought the spatial information of target observed is included in high frequency components of wavelet coefficients, while spectral information of target observed is contained in low frequency components of wavelet coefficients. The fusion strategy is the low frequency component of multi-spectral images is substituted by the high frequency component of panchromatic image to inject spatial details of panchromatic image and at last high spatial resolution multi-spectral images are produced. So we could utilize the least square principle to estimate the optimal value of spatial information to complete fusion making images'prior information-the ratio of spatial resolution as the ratio of the observation precision. Fusion result evaluation shows the proposed method could integrate spatial information of panchromatic image and spectral information of multi-spectral images better than other methods and produce more ideal high spatial resolution multispectral images.(3) Since a huge amount of remote sensing data, the computational efficiency is particularly important for image fusion. DCT (Discrete Cosine Transform) is a special case of Fourier transform. The operation speed is higher than the wavelet transform and has been applied to multi-spectral image and panchromatic image fusion. However, integration of existing methods failed to conduct in-depth analysis model so as to it lack theoretical basis. In this paper, a new method is proposed based on the ratio of fusion image spatial resolution combined with DCT-block mode to fuse multi-spectral images and panchromatic image. Final results show that the abilities of spectral characters preserved of method proposed is as good as the method based on wavelet transform and ideal fusion result is obtained, but its operating efficiency is enhanced simultaneously. (4) SAR image is different with visible light remote sensing image completely, reflecting the backscattering properties of ground target about microwave. SAR could observations earth in all-time, all-weather, penetrate clouds, rain, snow, fog and some Surface materials of earth, information of SAR image and visible light image are a kind of complementary relationship. Affected by speckle noises, the interpretation of SAR image is difficult, it is necessary to fuse SAR image and visual light image to hence SAR image's interpretation. To avoid loss of SAR image in previous fusion method based on wavelet transform. A new approach is proposed based on atrous wavelet and generalized IHS transform. It could control the amount of visual light image's information by changing the threshold value according to the need for interpretation. Experiments show that the method proposed could fuse SAR image and visual light image better than other method based on wavelet transform according different needs, and retains all information of SAR image. It has more freedom in practice.(5) Remote sensing image fusion method based on multi-scale analysis is a research hotspot in recent years. Contourlet transform is further development of wavelet transform, it extends the good characteristics of wavelet from one-dimensional space to two-dimensional space, overcomes the shortcoming of wavelet in two-dimensional space that could not capture, represent linear spatial entities efficiently. Contourlet transform has good multi-resolution, localized and directional characteristics. There are more choices when Contourlet transform is used to fuse images, and the fusion results will be more delicate. To improve the ability of SAR image interpretation, the method based on Contourlet+IHS transform is proposed to fuse SAR image and multi-spectral images. The strategy is low frequency coefficients of SAR images is hold to maintain the spectral characteristics of SAR images, in high frequency domain, the coefficients of absolute maximum is selected to improve gap between SAR image's coefficient and multi-spectral coefficient, the fusion result will be more clear. The experiment results that are evaluated by visually and statistically shows that more successful results is achieved in the fusion of SAR and Multi-spectral images using the method proposed in this paper than methods based on IHS and DWT+IHS transform. The color information of multi-spectral images and the spatial details of SAR image are integrated very well, and the level of interpretation of sar image is increased significantly relative to original SAR image.