Study On Thin Cloud Removal In Remote Sensing Images Based On Generative Adversarial Network

With the development of satellite remote sensing technology,various spaceborne sensors provide sufficient data support for industries and mass applications such as geographic and national conditions monitoring,agricultural conditions monitoring,safety warning,positioning and navigation.And optical remote sensing satellite has received more and more attention due to its high temporal,spatial resolution and rich surface coverage information.The application demand of optical remote sensing satellite is increasing day by day.However,when there are clouds in the atmosphere,the process of collecting image data will be affected,and the information of ground features will be blocked,which will reduce the quality of optical remote sensing image data,thereby affecting the industrial application of data.Therefore,studying the mechanism of cloud interference in optical remote sensing images and the restoration of lowquality images affected by clouds has strong practical significance.The research work of this article is as follows:(1)Analyze the composition of the atmosphere and the formation process of thin clouds,introduce the spatial frequency domain characteristics,spectral characteristics and time characteristics of thin clouds in remote sensing images.Compare and analyze differences exist in each band when the multispectral remote sensing images are blocked by thin clouds.Proposed the research idea of selecting the near-infrared and short-wave infrared band data that are less affected by thin clouds,and these bands can assist in the restoration of remote sensing images occluded by thin clouds.(2)Aiming at the problem of lacking feature information in the area covered by thin clouds,the proposed method of removing thin clouds from remote sensing images based on the generative adversarial network.At the same time,the loss function including the color parameters is designed to guide optimization of model parameters to improve the effect of thin cloud removal.(3)Aiming at the unnatural transition problem in the results of remote sensing image cloud removal,this artical studied the thin cloud removal method of remote sensing image based on attention generative adversarial network.By simulating the thin cloud formation in remote sensing images as the process of clear sky image and thin cloud superposition.Research how to improve the accurate extraction of discrete thin cloud regions by the attention network model.At the same time,study how to use the cirrus cloud band to assist the attention network to generate attention weight images,and use the attention images to guide the generative adversarial network to process the thin cloud area,improving the problem of unnatural transitions in the repair results.Finally,use the model to process real thin cloud image data,comparing with traditional methods,and analyze the advantages and disadvantages of this method.The data corresponding to the processing results of the improved method are: the peak signal-to-noise ratio value is 25.5801,the structure similarity value is 0.703135,and the mean square error value is 0.002916,which is better than the network before the improvement and other methods.The results show that the thin cloud removal method based on attention generation adversarial network can effectively remove thin clouds and supplements the image texture information.And the improved images is natural and have good integrity.