[R-sig-Geo] Random forest for forest cover classification

[Juan Pablo Carranza]

Dear Enoch, I strongly recomend to try an segmentation approach instead a
classification. However, here's some code you can use:

library(rgdal)
library(raster)
library(caret)
library(kerndwd)
library(sp)
library(RColorBrewer)
library(dplyr)
library(leaflet)
library(sf)
img <- brick("******.tiff")
names(img) <- paste0("B", c(1:*))
plot(img)
library(sf)
sample <- st_read("******gpkg")
sample  = as(sample, 'Spatial')
sample$class = as.factor(sample$class) ; responseCol <- "class"
pixels = data.frame(matrix(vector(), nrow = 0, ncol = length(names(img)) +
1))

# A function to overlap de sample polygons on the image.
for (i in 1:length(unique(sample[[responseCol]]))){
  category <- unique(sample[[responseCol]])[i]
  categorymap <- sample[sample[[responseCol]] == category,]
  dataSet <- extract(img, categorymap)
  dataSet <- dataSet[!unlist(lapply(dataSet, is.null))]
  if(is(sample, "SpatialPointsDataFrame")){
    dataSet <- cbind(dataSet, class = as.numeric(category))
    pixeles <- rbind(pixeles, dataSet)
  }
  if(is(sample, "SpatialPolygonsDataFrame")){
    dataSet <- lapply(dataSet, function(x){cbind(x, class =
as.numeric(rep(category, nrow(x))))})
    df <- do.call("rbind", dataSet)
    pixels <- rbind(pixeles, df)
  }
}

# Define a number of pixels to use in the training of the model. More
pixels, more accuracy but more computational time.
n <- 1000

# Take a subsample in order to reduce computational times.
spixeles <- pixeles[sample(1:nrow(pixeles), n ), ]

# Train model.
modFit_rf <- train(as.factor(class) ~ ., method = "nnet", data = spixeles)
modFit_rf

# Predict.
beginCluster()
preds_rf <- clusterR(img, raster::predict, args = list(model = modFit_rf))
endCluster()

# Mapping results
maxpixels =  1e+06
m <- leaflet() %>%
  addProviderTiles('Esri.WorldImagery') %>%
  addRasterImage(preds_rf , opacity = 0.6) %>%
  addScaleBar(position = "topright") %>%
  addMiniMap( position = "bottomleft")
m

# Export raster
writeRaster(preds_rf, "estimacion.tif")



*Juan Pablo Carranza*
Mgter. en Administración Pública
Lic. en Economía




El mar., 13 ago. 2019 a las 5:50, Enoch Gyamfi Ampadu (<egampadu using gmail.com>)
escribió:

> Dear List,
>
> Please I am now learning R to apply to my analysis. I am doing a forest
> cover classification using Random forest. So far I have stacked the image,
> done atmospheric correction, and clipped out my study area. I have created
> some training sites of the cover classes in ArcMap and imported into R as
> well as the clipped image of my study sites. I have been doing try and
> error with some online codes. I will be glad if you could provide me with
> some codes to apply.
>
> Thank you.
>
> Best regards,
>
> Enoch
>
> --
> *Enoch Gyamfi - Ampadu*
>
> *Geography & Environmental Sciences*
>
> *College of Agriculture, Engineering & Science*
>
> *University of KwaZulu-Natal, Westville Campus*
>
> *Private Bag X54001*
> *Durban, South Africa **– 4000**.*
> *Phone: +27 835 828255*
>
> *email: egampadu using gmail.com <egampadu using gmail.com>*
>
>
> *skype: enoch.ampadu*
> *The highest evidence of nobility is self-control*.
>
> *A simple act of kindness creates an endless ripple*.
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo using r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

	[[alternative HTML version deleted]]