Research On Super-resolution Reconstruction Algorithm Of Remote Sensing Image

With the rapid development of space technology and the promotion of practical application,the development of micro-nanosatellites has made great progress.With the advantages of small size,light weight,low power consumption,short development period and low cost to complete complex space test missions,micro and nano-satellites play an increasingly important role in the fields of national defense,scientific research and commercial use.Because of the limitation of micro-and nano-satellite hardware equipment,it is a contradiction between improving spatial resolution and hardware equipment.How to obtain high resolution remote sensing images on limited hardware has became the direction of researchers.Image super resolution(SR)reconstruction is a process in which one or more low-resolution images by means of post-processing,i.e.signal processing or image processing to obtain high-resolution images,they are not changing the premise of the original hardware equipment.Traditional reconstruction algorithms often have the problems of long reconstruction time and poor robustness.However,with the application of convolution neural network to image super-resolution reconstruction,these problems can be avoided.However,the existing shallow structure convolution neural network reconstruction algorithms have good reconstruction effect for the data of simple image content and weak constraint.The content structure of the image is complex when the actual remote sensing image is obtained,and the existing network model will expose the problem of lack of representation ability.Therefore,in order to obtain the reconstruction model of remote sensing image,this paper deeply studies the super-resolution reconstruction algorithm of remote sensing image based on convolution neural network.The main content and innovation of this paper are as follows:(1)The feasibility of super-resolution reconstruction of single frame remote sensing image is analyzed,and the importance of super-resolution reconstruction in remote sensing image processing is explained.Then the super-resolution reconstruction is introduced and the related content of convolution neural network is elicited.The superiority of convolution neural network algorithm in image super-resolution reconstruction can be obtained from the relevant theoretical knowledge.(2)A super-resolution reconstruction structure based on convolution neural network is proposed for remote sensing images,which uses the end-to-end super-resolution framework of the convolution neural network.In the network structure,the feature maps are extracted by different convolution,then the feature maps are combined to extract the features again by nonlinear mapping.The advantage of this approach is that the image features are fully extracted by using the diversity of convolution cores.The experimental results show that theproposed network structure is of good quality for the reconstruction of single remote sensing image,and the convergence rate of the training stage is also improved.By using PSNR(Peak Signal-to-Noise Ratio)and SSIM(Structural Similarity Index)to quantitatively analyze the reconstruction results.(3)In order to meet the requirements of different magnification in practical applications,a PC-based software system is developed and verified by remote sensing image reconstruction of different scenes.