train: Train models for forest detection

In deforestable: Classify RGB Images into Forest or Non-Forest

Train models for forest detection

Description

As input data, the function needs two folders- Nonforestdir with images of non-forest and Forestdir with ones of forest. train() uses all images in both folders to train a model. Putting an image into an incorrect folder is equivalent to tagging the image incorrectly.

Usage

train(
  n_pts,
  model = c("fr_Non-Param", "fr_Param"),
  Forestdir,
  Nonforestdir,
  train_method = c("cv", "train"),
  k_folds,
  parallel = FALSE
)

Arguments

n_pts	matters only when train_method='cv'. Defines the size of the square sub-frames into which images would be split during cross-validation.
model	which model to train
Forestdir	path to the directory with (only) forest images
Nonforestdir	path to the directory with (only) non-forest images
train_method	how to train the model: simple training, cross-validation.
k_folds	matters only when train_method='cv'. The number of folds in the k-fold cross-validation setup.
parallel	matters only when train_method='cv'. Boolean. whether or not use a parallel setting during cross-validation

Details

Currently, both fr_Non-Param and fr_Param use parameter n_pts only in the testing part of cross-validation, not during training. Training is always done on whole original images in the training folders.

Value

object of class ForestTrain potentially with a sub-class. See Class_ForestTrain.

Examples

library(deforestable)
n_pts <- 20

# Choosing folders with training data
Forestdir <- system.file('extdata/Forest/', package = "deforestable")
Nonforestdir <- system.file('extdata/Non-forest/', package = "deforestable")

k_folds=3;

#### Read the target image ####
tg_dir <- system.file('extdata/', package = "deforestable")
test_image <- read_data_raster('smpl_1.jpeg', dir = tg_dir)


#### Models ####


# Simple training of the non-parametric model
Model_nonP_tr <- train(model='fr_Non-Param', Forestdir=Forestdir, Nonforestdir=Nonforestdir,
                       train_method='train', parallel=FALSE)

res <- classify(data=test_image, Model=Model_nonP_tr,
                n_pts=n_pts, parallel=FALSE, progress = 'text')

tmp_d <- tempdir(); tmp_d
jpeg::writeJPEG(image=res, target=paste(tmp_d,'Model_nonP_tr.jpeg', sep='/'))


# Cross-validation of the non-parametric model
Model_nonP_cv <- train(n_pts=n_pts, model='fr_Non-Param', Forestdir=Forestdir,
                       Nonforestdir=Nonforestdir, train_method='cv',
                       k_folds=k_folds, parallel=FALSE)

res <- classify(data=test_image, Model=Model_nonP_cv,
                n_pts=n_pts, parallel=FALSE, progress = 'text')

tmp_d <- tempdir(); tmp_d
jpeg::writeJPEG(image=res, target=paste(tmp_d,'Model_nonP_cv.jpeg', sep='/'))




# Cross-validation of the parametric model
Model_P_cv <- train(n_pts=n_pts, model='fr_Param', Forestdir=Forestdir,
                    Nonforestdir=Nonforestdir, train_method='cv',
                    k_folds=k_folds, parallel=FALSE)

res <- classify(data=test_image, Model=Model_P_cv,
                n_pts=n_pts, parallel=FALSE, progress = 'text')

tmp_d <- tempdir(); tmp_d
jpeg::writeJPEG(image=res, target=paste(tmp_d,'Model_P_cv.jpeg', sep='/'))


# Simple training of the parametric model
Model_P_tr <- train(model='fr_Param', Forestdir=Forestdir, Nonforestdir=Nonforestdir,
                    train_method='train', parallel=FALSE)
res <- classify(data=test_image, Model=Model_P_tr,
                n_pts=n_pts, parallel=FALSE, progress = 'text')

tmp_d <- tempdir(); tmp_d
jpeg::writeJPEG(image=res, target=paste(tmp_d,'Model_P_tr.jpeg', sep='/'))

deforestable documentation built on Oct. 16, 2022, 1:06 a.m.