downscaledl-package: Downscale of RS Images using Deep Learning

In downscaledl: Downscale of RS Images using Deep Learning

Description

Downscaling the coarse-resolution remote sensing images into fined-resolution images is important for preprocessing. This package provides the functions for this with high accuracy. Currently, the resautonet is used as the deep learning algorithm for downscaling. This work was done based on this paper, Lianfa Li (2019) <doi:10.3390/rs11111378>.

Details

This package provides the downscaling functions for the coarse-resolution remote sensing images. Currently, we used the deep learning algorithm as the base learner in downscaling.

The major functions defined in this package include:

AutoEncoderModel:The residual network function used in downscaling;

ResautoDownscale:The major downscaling function that uses AutoEncoderModel;

r2_squ:The function calculating R2 ;

rmse:The function calculate RMSE.

Author(s)

Lianfa Li

Maintainer: Lianfa Li <[e-mail]>

References

~~ Literature or other references for background information ~~

Examples

  #Load the high-resolution raster of a covariate, 
  # elevation to be used for downscaling  
  eleFile=file.path(system.file(package = "downscaledl"), "extdata", "sample_ele.tif")
  ele=raster::raster(eleFile)

  #Load the coarse-resolution raster of the target variable to be downscaled  
  coarseFile=file.path(system.file(package = "downscaledl"), "extdata", "sample_coarse_res.tif")
  coarseRst=raster::raster(coarseFile)

  #Extract x and y to be used as two predictors in downscaling 
  xRast=ele
  yRast=ele
  pos=raster::rasterToPoints(ele,spatial=TRUE)
  cell=raster::cellFromXY(xRast,pos)
  xyc=sp::coordinates(pos)
  xRast[cell]=xyc[,"x"]
  yRast[cell]=xyc[,"y"]

  #Merge the covariates 
  covStk=raster::stack(xRast,yRast,ele)
  names(covStk)=c("x","y","ele")

  #Use the fine-resolution covariate (elevation) as the target image 
  fineTarget=ele

  #Set the paramneters and start to downscale ...  
  ares=ResautoDownscale(covStk,fineTarget,coarseRst,ss= 0.2, cores= 5, thresh = 0.01,ntime=3)
  
  #Show the iteration results 
  message(paste(capture.output(ares$diogRMSE), collapse = "\n",sep=""))
  
  #Show the optimal results in the final predictions 
  message(paste("test R2:",round(ares$r2,2),",test RMSE:",round(ares$rmse,4),sep=""))  
  
  #Obtain the downscaled
  downscaled_img=ares$raster
  
  #Save the current par setting 
  curpar = par(no.readonly = TRUE)
  
  #Set the new par setting 
  par(mfrow=c(1,2),mar=c(1,1,1,1))
  
  #Show the final predictions of fine resolution and 
  # original coarse-resolution image for a comparison 
  raster::plot(coarseRst)
  raster::plot(downscaled_img)
  
  #Restore the previous par setting 
  par(curpar) 

downscaledl documentation built on Oct. 30, 2019, 11:35 a.m.