Remote Sensing Image Fusion Based On Sparse Representation

With the rapid development of remote sensing technology,satellite sensors have acquired a large number of remote sensing images with different spatial resolution and spectral resolution.These images are widely used in many fields such as forest fire detection,smart city and weather forecasting,which facilitate people's lives and also promote social development.However,remote sensing images acquired based on optical systems are contradictory between the spatial resolution and the spectral resolution.At a fixed signal-to-noise ratio,the spectral resolution increases at the expense of the spatial resolution,and therefore we can only get multispectral images with low spatial resolution but high spectral resolution and panchromatic images with less spectral information but rich spatial detail.Many remote sensing applications need to combine the advantages of both multispectral image and panchromatic image to generate a multi-spectral image with high spatial resolution to meet the needs of further applicationbased on the image,and remote sensing image fusion is an effective way to solve this problem.Information expression based on sparse representation is very hot in information processing field.Sparse representation aims to find suitable bases in non-orthogonal base system to expresses the signal and it has been widely used in the field of signal acquisition and compression.Recently,many scholars introduce sparse representation into the field of remote sensing image fusion and improve it with the professional knowledge of image fusion,these methods can provide better spectral quality and superior spatial information in the fused image than its counterparts.Based on the research of sparse representation theory,this paper makes a futher research on the fusion of remote sensing image based on sparse representation.The main work is as follows:1.Analyze the process of over-complete dictionary trained by traditional K-SVD method,in order to overcome the shortcomings of its large amount of training and time-consuming,K-means clustering is introduced in the preprocessing of training samples.We select the appropriate features to cluster the training samples and reduce the samples with higher similarity,by this way we reduce the training amount while keeping the diversity of the training samples,thus improving the efficiency of dictionary training.2.The exist of similar dictionary atoms and less used dictionary atoms reduce the performance of the dictionary,thus the normalized panchromatic image blocks are used to replace them to obtain an adaptive dictionary,which can more accurately sparsely represent multispectral image and panchromatic image.3.We use the adaptive dictionary to sparsely represent the intensity component of multispectral image and panchromatic image,and then choose suitable rules to fuse them.Due to the importance of maximum sparse coefficient,we decompose the sparse coefficient into maximum sparse coefficient and residual sparse coefficient,and then fuse them with different fusion rules.The experimental results show that the fused images obtained by this method have better subjective visual effects and better objective evaluation indexes.