Research On Image Super-resolution And Fusion Algorithms Based On Conditional Generative Adversarial Learning

In human life,images are one of the most important information carriers.However,in the process of acquiring images,it is easily affected by external degradation factors and the device itself,which results in the acquired images not meeting the actual needs.Therefore,how to restore low-resolution images to corresponding high-resolution images through superresolution reconstruction technology has been a problem that researchers have been working on to solve.With the continuous development of artificial intelligence technology,superresolution reconstruction technology has been widely used in remote sensing imaging,image compression,medical imaging and public security and other fields.In this paper,combined with the characteristics of deep learning,the conditional generative adversarial network is used for super-resolution reconstruction,and the method is improved and extended to the remote sensing spatio-temporal fusion technology.The specific work is as follows:(1)Image super-resolution based on conditional generative adversarial networkWith the development of deep learning,in recent years,generative adversarial networks have made great progress in image super-resolution technology.However,due to the instability and high distortion of the traditional generative adversarial network,this paper proposes an image super-resolution algorithm based on the conditional generative adversarial network,which is designed to use high-resolution real images as condition variables.To better reconstruct image with a large-scale factor,this paper proposes an enhanced Laplacian pyramid network as a generator model of the conditional generative adversarial network,which progressively reconstructs high-resolution images at multiple pyramid levels.The proposed method fuses low-and high-level features for the residual image learning achieves better generalization than those only based on high-level information.In addition,this paper also improves the loss function,optimizes the network by combining multiple loss functions,including the adversarial loss function,the VGG loss function and robust Charbonnier loss function to obtain high-quality results.(2)Spatio-temporal fusion based on conditional generative adversarial networkConsidering that remote sensing images also have a resolution gap problem,the above algorithm is applied to the remote sensing spatio-temporal fusion technology and improved,and a spatio-temporal fusion method based on conditional generative adversarial network is proposed.To improve the calculation efficiency of the generator network,this paper designs an asymmetric Laplacian pyramid network,in which a lightweight residual block is used in the internal structure of the network,which can reconstruct a large-scale factor image,and solve the problem of under-fitting model.Since there are differences in satellite parameters,imaging environment and atmosphere between different sensors,the correspondence between different sensor images is established by learning two conditional generative adversarial network models.Through actual data testing,the results show that the algorithm proposed in this paper can effectively improve the spatio-temporal fusion effect of remote sensing images.