Key Techniques And Methods For Constructing Long Time Series Remote Sensing Image Data

Compared with the single image,the time-series image has a time dimension,which contains much more information than a single image which contains only spatial dimension.Through the remote sensing images of time series,we can get the characteristic changes of region over time and,which is able to reflect the information of the region.Therefore,time series images become an important information which can not be obtained in the terms of urban expansion,forest degradation,global warming and ecosystem destruction.In the process of construction of the remote sensing images of time series,there are three major problems:the removal of image cloud,the generation of missing image,the precision evaluation of image fusion.This paper will coduct a study on it and put forward the corresponding improvement methods in order to improve the quality of time series images and construct the remote sensing imagesof time series with medium resolution in Fuzhou.The main research contents and results are as follows:(1)The improving of Criminisi method.The cloud cover is a significant cause which lead to the lack of information of the remote sensing images with medium resolution.According to statistics conducted by UGUS,the average cloud of Landsat images is 35%,so the top priority is to take small amout of cloud removing into consideration.Therefore,on the basis of the Criminisi model,this paper adjusts the weight strategy of the priority calculation and introduces the structure information of the auxiliary image in the best match calculation.The experimental results show that the improved Criminisi method is able to improve the accuracy of cloud filling and get better cloud removal effect.(2)The research on 3DSKRFM.The STARFM model uses a fixed size window to find similar pixels,and searches similar pixeluses by the method of isotropy without taking into account actual characteristics,such as the image anisotropy and the differencesf texture complexity.So this paper is going to use control optimization of kernel weight function,which can makes the window size and shape would be adaptive with pixel and local field correlation,and takes into further acount the relevance of different bands of remote sensing images,extends steering kernel to 3D,so as to develop a three-dimensional adaptively local steering kernel regression fusion model(3DSKRFM).Compared with the other two spatio-temporal fusion models,the experimental results show that 3DSKRFM model not only possesses the best fusion effect but also a stronger capacity to resist noise.(3)The research on the fusion image of precision evaluation based on student-t test.For the existing quality evaluation index is not based on the entire image and can not evaluate objects,this paper introduces the test method and puts forward a model of visual evaluation based on the spatial distribution to improve the accuracy and scientificity of evaluation.(4)The construction of the remote sensing images of time series with medium resolution in Fuzhou.According to the existing data,remote sension data with 30 meters spatial resolution of the four seasons of FuZhou country from 2001 to 2015 were established through the cloud removal,image fusion and other steps.