Research On Remote Sensing Image Super-resolution Technology Based On Adversarial Generative Network

In the imaging system,due to the influence of various degradation factors during the transmission process of the image signal,the imaging quality will be affected to different degrees and the image resolution will be reduced.The image super-resolution reconstruction technique can be used to recover super-resolution images from lowresolution images,and this technique is widely used in many scenarios.This paper is based on the research in the field of super-resolution of remote sensing images,with adversarial generative networks as the main network framework,and the overall research content and innovative work are as follows:Firstly,a remote sensing image super-resolution technique based on cascading residual adversarial networks is proposed.The network is based on SRGAN,which mainly improves the structure of the generator sub-network,using the cascading residual blocks as the core of the generator module.The extracted low-frequency image features are grouped and convolved before entering the residual block to reduce the number of parameters,while the batch standard layer within the original residual block is removed,and then the features of the different cascade blocks are aggregated in a cascade manner before being upsampled to complete the image reconstruction process in the generator.Secondly,the multi-scale residual generative adversarial networks algorithm for super-resolution reconstruction of remote sensing images is proposed.The generator module of the algorithm uses a multi-scale residual block as a core structure that makes full use of feature information at different scales.The parallel information of the multiscale features is first extracted to enable full interaction between information streams by means of selective kernel fusion,followed by spatial and channel attention mechanisms for capturing contextual information,and finally the process of aggregating multi-scale features based on attention mechanisms is implemented.The discriminator module introduces the concept of relative discriminators,enabling the network to learn clearer edges and more detailed information.Finally,the Py Qt5 development tool is used to design and implement the user interface for the super-resolution of remote sensing images to visualise and automate the super-resolution reconstruction.