rf_image_classification.rsx

##img=raster
##trainData=vector
##responseCol=field trainData
##nsamples=number 1000
##output=output raster
library(sp)
library(raster)
library(caret)

names(img) <- paste0("B", c(1:5, 7))

dfAll = data.frame(matrix(vector(), 0, length(names(img)) + 1))

for (i in 1:length(unique(trainData[[responseCol]]))){
  category <- unique(trainData[[responseCol]])[i]
  categorymap <- trainData[trainData[[responseCol]] == category,]
  dataSet <- extract(img, categorymap)
  dataSet <- dataSet[!unlist(lapply(dataSet, is.null))]
  dataSet <- lapply(dataSet, function(x){cbind(x, class = as.numeric(rep(category, nrow(x))))})
  df <- do.call("rbind", dataSet)
  dfAll <- rbind(dfAll, df)
}

sdfAll <- dfAll[sample(1:nrow(dfAll), nsamples), ]

modFit_rf <- train(as.factor(class) ~ B3 + B4 + B5, method = "rf", data = sdfAll)

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

output = preds_rf

Thank you for this post

Hello, I am doing my M.Sc. thesis with Sentinel 2 dataset. I want to do random forest classification by running your code in QGIS. But the problem is I dont understand the meaning of this line:
c(paste0("B", 1:5, coll = ""), "B7").
Can you help me?

I have just edit the code like following:
##img=raster
##trainData=vector
##responseCol=field trainData
##nsamples=number 1000
##output=output raster
library(sp)
library(raster)
library(caret)

names(img) <- c(paste0("B", 2:8, coll = ""), "B8A","B11", "B12")

dfAll = data.frame(matrix(vector(), 0, length(names(img)) + 1))

for (i in 1:length(unique(trainData[[responseCol]]))){
category <- unique(trainData[[responseCol]])[i]
categorymap <- trainData[trainData[[responseCol]] == category,]
dataSet <- extract(img, categorymap)
dataSet <- lapply(dataSet, function(x){cbind(x, class = as.numeric(rep(category, nrow(x))))})
df <- do.call("rbind", dataSet)
dfAll <- rbind(dfAll, df)
}

sdfAll <- subset(dfAll[sample(1:nrow(dfAll), nsamples), ])

modFit_rf <- train(as.factor(class) ~., method = "rf", data = sdfAll)

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

output = preds_rf
But, after running I have a message the output file does not created correctly.

Why it does not specify the number of trees?

Hello.

Sorry for the message but I see the error.

Now, I just don't understand why the input is a raster (multispectral) and then is interprets the bands, because in my case it gives an error in "names(img) <- paste0("B", c(1:5, 7))" and I put the number of bands that I have in the composite image. However I always get an error in this line. Is the name that is wrong?