Research On Sparse Theory Based The High Temporal And High Spatial Resolution Remote Sensing Image Simulation

In recent years,due to the increasing demands of dynamic observation,launched many satellites with a variety of sensors.However,considering the economic reasons such as few both high spatial resolution and high time resolution of the sensor.In order to simultaneously monitor to a wider area and better observation details,this article is based on a pair of remote sensing image,image fusion based on space-time model,realize high temporal high spatial resolution image simulation.This method first by image super-resolution technology and low altitude image information will be high temporal resolution low spatial resolution image data(when the low-level image)rebuilt,thus improve the spatial resolution image;On this basis,the model based on highpass,the low spatial resolution and high time resolution image(low altitude)image data fusion to get the final result.Spatial resolution in order to reduce the difference is too large,the error of the whole process is divided into two parts:the first stage super resolution reconstruction of sparse representation technology improve when the low-level image spatial resolution by improving the dictionary training methods,training a priori images;The second stage,the use of the known high altitude low spatial resolution image and has improved when the low-level image by highpass model fusion,through improved bilateral filter is sensitive to the airspace and domain sensitive parameters,get a better image denoising image effect,high altitude when image simulation.For the need to monitor the area and need to change the details of the changes in land monitoring is of great significance is very important.In this paper,the main work is divided into the following several parts:first,spatiotemporal algorithm based on improved single dictionary learning algorithms of the images,aimed at the problem of low efficiency of dictionary learning training,update the improved K-SVD algorithm used in dictionary stage,and then improve the efficiency of the dictionary training and the reconstructed image quality.Second,in time and space algorithm based on improved single dictionary learning algorithms of the images,through the dictionary training before joining MNF transform,reduce the amount of data when the training dictionary,reduced the noise,get better image effect.Image simulation algorithm in this paper with a pair of transcendental image,under the condition of lack of transcendental image,obvious advantages,through comparative tests prove that the algorithm is feasible.Three,through improved bilateral filter domain and airspace parameters,makes the image denoising effect is better,to make it with the actual image' closer to the vegetation change.In this paper,the high temporal Resolution of low spatial Resolution image data selected Terra Moderate Resolution Imaging Spectroradiometer(MODIS)data,high spatial Resolution and low time Resolution image is the environment a number of small satellite star B CCD data,using 2014.05.29 and 2014.07.17 MODIS data,as well as 2014.05.29 HJ CCD data,2014.07.17 CCD data simulation.The experimental results show that the image of the model prediction with high temporal resolution MODIS images and HJ images of high spatial resolution characteristics.Similarity to a maximum of more than 0.8.Compared with other space-time fusion model algorithm,this algorithm has better convergence effect and requires less known image,in the case of rarely known image is a very favorable conditions.