Multi-source Image Fusion Algorithm Based On Convolutional Neural Network And Super-resolution

Image fusion is the process that fuses two or more information-dispersed multi-source images of the same scene into an image with more details and spectral.The phenomena appear such as spectral and detailed information scattered in the shoot due to the interference of various factors such as the type of imaging sensor,the focusing direction,and the location where the scene is shotted.Image fusion algorithm is usually used to fuse the multi-source images to solve these problems and get a more accurate and reliable fusion image.This paper focuses on the fusion of remote sensing image and multi-focus image.As the traditional fusion algorithms lack of detail easily and the other improved algorithms are effect but often take a long time,the shorter and better fusion algorithms are proposed to make a better fusion process.The main contents and innovations of this paper include the following aspects:1.In this paper,an improved multi-focus image fusion method based on convolutional neural network is proposed according to the traditional fusion algorithms.The feature images of the source image in more directions are obtained by increasing the number of convolution filters in the network model.First,the back propagation is used to train the network model,the input image is approximation to the input in the network model;Second,the source images are decomposed into a set of feature maps by the trained network model,and the feature maps are fused with the rule of region image definition,the fusion image obtained with the fused feature maps reconstructed by convolution filters.The multi-focus image fusion comparison experiment results show that the fusion result of the algorithm not only improves the evaluation index,but also has better visual effect and shorter operating time.2.An improved remote sensing image fusion method based on super-resolution is proposed to make full use of spatial information of multispectral image and obtain better fusion quality.First,the original mulispectral image is reconstructed by Accelerating the Super-Resolution Convolutional Neural Network(FSRCNN),which enhance the spatial information while expanding its size.Then the expanded mulispectral image is transformed with IHS transform and got the I componen.the panchromatic image and the reconstructed I component of mulispectral image are fused with the fusion method based on wavelet transform,the fusion rule is theabsolute value and change the high frequency components of the fusion image are all derived from the panchromatic image in the traditional algorithms.Finally,the fused multispectral image is obtained via inverse IHS transform.The experiment results show that the algorithm outperforms compared with other algorithms,and reduces the loss of spatial information and spectral information in the process of image fusion effectively.