Research On SAR And Visible Image Fusion Algorithm

Since entering the age of spaceflight,satellite remote sensing has an increasing market share in the fields of surveying and mapping,infrastructure,transportation,agriculture,forestry and animal husbandry.Synthetic Aperture Radar(SAR)sensors based on active microwave imaging mechanism are used for all-weather imaging.The penetrability of images to soil and water bodies and the characteristics of disregarding the interference of climatic conditions make the processing of SAR data more and more important.Optical satellite images based on light reflection have rich spectral information and occupy the mainstream position in various conventional monitoring fields,but are easily interfered by external conditions.This article takes SAR and visible light images as the research object,and conducts research from the field of multi-source image fusion.It is concluded that the SAR and visible light fusion algorithm still has shortcomings,and the fusion image has the problems of spectral distortion and loss of detailed information.In-depth research on multi-source image fusion algorithm,the main research content is divided into the following aspects.(1)Summarize the shortcomings of the current multi-source image fusion algorithm and the direction of improvement,and summarize the advantages of the convolutional neural network for multi-source image fusion;Analyze the performance requirements of SAR and visible light fusion images,and select a joint evaluation method of subjective vision and suitable objective indicators to comprehensively evaluate the fusion image;According to the characteristics of SAR and visible light images,the fusion preprocessing process of SAR and visible light images is summarized.(2)The image super-resolution reconstruction algorithm replaces the interpolation method for resampling visible images,By comparing the super-resolution reconstruction algorithms of different models,a convolutional neural network image super-resolution reconstruction(IRCNN)model combining optimized models and discriminative learning was selected for super-resolution image enhancement of visible light images.The IRCNN model is improved based on the characteristics of large amount of information and detailed information of remote sensing images.The image enhancement results show that the output image based on the improved IRCNN model is superior to other super-resolution models in edge retention and overall image clarity.(3)A fusion algorithm of SAR and visible light image based on convolutional neural network is proposed.The algorithm uses the powerful feature extraction capabilities of convolutional neural networks to fully extract the effective information on SAR and visible light images.Then,a fusion strategy with maximum regional energy is used to fuse the effective information in the SAR and visible light images.Compare the output images of this algorithm with the other 6 fusion algorithms,and analyze the output images of each fusion algorithm through objective indicators and subjective vision.The results show that the algorithm in this paper has a good performance in image spectral retention and details and texture information.