Remote Sensing Image Super Resolution Method And Its Application In Water Monitoring

Remote sensing technology has the advantage of quickly acquiring large-scale data in a non-contact manner.It is the main means of water resources and water environment survey and water body information extraction.The resolution of remote sensing images reflects the quality of the data and whether it can provide more and more accurate information.Due to external environmental interference and sensor manufacturing process limitations,remote sensing image quality has different degrees of decline.Super-resolution reconstruction technology has been widely studied as an efficient and low-cost method to improve image quality.In the existing super-resolution reconstruction method,the deep learning method is regarded as a "new rising star" compared to the conventional super-resolution reconstruction methods such as interpolation,statistical methods,signal decomposition,etc.,Because of its simple implementation process and the high reconstruction quality,it is favored by scholars in the field of super-resolution.In this paper,a convolution-deconvolution alternating neural network(CDANN)is proposed based on the difference between remote sensing image and natural image.The network adopts the 3D convolution/deconvolution layer,and repeatedly extracts the advanced and abstract feature information of the image through the concatenated convolution layer and deconvolution layer,and introduces the residual learning to simplify the network.Considering the remote sensing images of the same scene,especially the difficulty in obtaining multi-spectral/hyperspectral remote sensing image data,the lack of data samples,and the large volume of data,CDANN's network structure design is simple and efficient.At the same time,the 3D convolution/deconvolution operation is used to extract the features of remote sensing image data,which can effectively protect the three-dimensional structural features of remote sensing image data,and take into account its spatial-spectral characteristics.In addition,the network defines the spatial loss function and the spectral loss function from the spatial and spectral perspective of remote sensing image data,so that the spatial and spectral information of the remote sensing image can be considered simultaneously in the network training process with a constant weight.Through experimental analysis,the proposed method can improve the quality of remote sensing image better than traditional methods and other deep learning methods.Combined with mixed pixel decomposition algorithm and super-resolution reconstruction,The accuracy of the water area change detection in Poyang Lake waters is effectively improved.In the end,The article is summary the research work of the thesis and further prospect the future work.