Remote Sensing Image Fusion Based On Bayesian Network And Evolutionary Algorithm

The existing high resolution satellites can provide two types of images:the multispectral (LM) image and the panchromatic (HP) image. The LM image is rich in spectrum information, but short of spatial information. The HP image has a high spatial resolution, but is lack of spectral information. Remote sensing image fusion is a technique that generates a high spatial resolution multi-spectral (HM) image by analyzing and fusing the spectral information contained in a LM image and the spatial information contained in a HP image according to a certain rule.In order to reasonably describe the correlational relationship of the LM image, the HP image and the HM image and solve the selection problem of the parameters, we propose a remote sensing image fusion method based on Bayesian Network and the Evolutionary Algorithm in this paper. The innovation points are as follows:(1) We propose a Pan-sharpening method based on Bayesian Network (BNP method). We use a directed acyclic graph to formulate the dependency relationship between the LM image M, the HP image P and the fused HM image F. We turn the fusion problem into solving a posterior probability p(F|M,P) and propose three rational assumptions about the conditional probabilities p(M|F), p(P|F), and the prior probability p(F).(2) We introduce EA to Pan-sharpening to select parameters automatically. The 12 parameters in the Pan-sharpening method are encoded as an individual vector in the EA. Two quality measurements are combined to form an objective function of the EA, which is used to evaluate the fitness value of the individual. The optimal parameters and the HM image are generated by optimizing the objective function.The new method is compared with some other methods using QuickBird data. The experimental results show that our method can generate a high-quality fused image, and the same parameters’values can be used for similar images.