Research On The Method Of Improving The Quality Of Luojia 1-01 Night-Light Remote Sensing Image

Night-light remote sensing Image by imaging environment,target radiation,the optical imaging system,electronic signal conversion and satellite platform flutter degradation factors,lead to "glow" observation images,cloud noise,and down sampling mass reduction phenomenon.The degradation phenomenon has largely restricted the later application.Therefore,on the basis of the existing data conditions,it is of great significance and value to use image processing algorithm to accurately and rapidly reduce or remove the influence of cloud in the Night-light remote sensing Image,and at the same time improve the spatial resolution and visualization effect of the image.The deep learning method based on neural network has the advantages of convenience,high efficiency,strong feature learning ability and good processing effect,which is the research hotspot of scholars.This paper takes the Luojia 1-01 image as the research object and studies the quality improvement of night-light remote sensing Image from the aspects of quality reduction model analysis,cloud removal and spatial resolution improvement.The research mainly involves the following three parts:Aiming at the problem that the expression of quality degradation model of nightlight remote sensing image is not detailed enough,this paper deduced and constructed a comprehensive quality degradation model of Luojia 1-01 image.Based on the principle of image quality degradation and the characteristics of image quality degradation,it is concluded that the main factors of image quality degradation of Luojia1-01 images are the interference of atmospheric components such as clouds and the influence of optical imaging system,and the main quality degradation is reflected by "glow",cloud noise and spatial resolution decline.Therefore,a comprehensive quality degradation model of Luojia 1-01 image was built based on the above quality degradation characteristics.The model can represent the comprehensive quality degradation process including "glow",cloud noise,and sampling degradation,which provides an important theoretical basis for the study of improving the quality of nightlight remote sensing image.Aiming at the problem that night-light remote sensing images are easily affected by clouds and fog in the imaging process,resulting in the phenomenon of cloud reflection and "glow" in the acquired images,this paper proposes a method of image dehazing based on the comprehensive degradation model of Luojia 1-01 image method.First design the RDCNN(Residual Dense Convolutional Neural Network)night-light images dehazing network,the network to RDCNN as the main body structure,using the CA(Channel Attention)and PA(Pixel Attention)module to improve the dehazing ability of the network.At the same time,the loss function is modified into a loss function with sparse constraints to improve the sharpness of processed images.Aiming at the problem that there is no night-light remote sensing datasets at present,an end-toend construction method of Luojia 1-01 image datasets were designed,and several groups of different types of quality reduction datasets were constructed for training network.The experimental results show that the images dehazing method proposed in this paper can effectively remove the cloud and fog from the night-light remote sensing image.After the processing,the cloud information of the image is obviously eliminated,other noise information is suppressed,the edge information of the image light is clearer and the image quality is improved significantly.Aiming at the problem of insufficient spatial resolution of night-light remote sensing image due to the limitation of optical imaging system,an end-to-end RDCNN based night-light remote sensing image super-resolution network was designed and constructed to further improve the resolution of night-light remote sensing image.The network inherits the advantages of the dehazing network in this paper and adds subpixel convolution as the upsampling part.In order to solve the problem of missing datasets of super-resolution of night-light remote sensing image,Luojia 1-01 image datasets was constructed for super-resolution reconstruction.Through several groups of experiments,it is proved that the method in this paper can effectively improve the spatial resolution of the image,and the noise of the image is suppressed and the edge information is clearer after the super-resolution reconstruction,which can improve the quality of the image and provide a better data source for the later application.The paper has 49 figures,12 tables,and 84 references.