PolygonSaturation: Polygon Saturation Identification Tool

In MBalthasar/superClassAnalysis: Toolbox for supervised classifications

Description

This function aims to identify the best number of polygons required per class within a training data set in order to get the highest accuracy results from a supervised classicifation. It furthermore enables to identify polygons which lead to lower classification accuracies for training data quality testing. The function will always use the same number of polygons per class and is therefore limited by the class with the least amount of polygons. It is overall recommended to have the same number of polygons for each class. The function will start with 2 polygons per class.

Usage

PolygonSaturation(img, model, trainData, valData, prodAcc, responseCol,
  overall, plot_graph, nSamples)

Arguments

img	A raster file.
model	The model which will be used for the classification. See train for model selection.
trainData	SpatialPolygonsDataFrame or SpatialPointsDataFrame containing the training locations.
valData	SpatialPolygonsDataFrame or SpatialPointsDataFrame containing the validation locations (optional). If no valData is given, the trainData will be split into 70 percent training and 30 percent validation.
prodAcc	TRUE or FALSE. If prodAcc is TRUE, the producer accuracy will be returned. If prodAcc is FALSE, the user accuracy will be returned.
responseCol	Character or integer giving the column in trainData, which contains the response variable.
overall	TRUE or FALSE. Defines, whether the overall accuracy should be included or not.
plot_graph	TRUE or FALSE. Defines, whether the resulting data.frame should be automatically plotted or not.
nSamples	Integer. Number of samples per land cover class.

Value

A data.frame with the accuracy numbers per class depending on the number of polygons used.

Examples

library(sp)
library(raster)
library(RStoolbox)
library(reshape2)
library(ggplot2)
library(randomForest)

# Load sample raster file
my_raster <- raster::brick(system.file(package = "superClassAnalysis", "extdata", "landsat_sample.tif"))

# Load sample training and validation data
my_train <- raster::shapefile(system.file(package = "superClassAnalysis", "extdata", "training_sample.shp"))
my_val <- raster::shapefile(system.file(package = "superClassAnalysis", "extdata", "validation_sample.shp"))

# Execute PolygonSaturation function
x = PolygonSaturation(img = my_raster, model = 'rf', trainData = my_train,
                     valData = my_val, nSamples = 100, responseCol = "class_name",
                     prodAcc = TRUE, overall = TRUE, plot_graph = TRUE)

MBalthasar/superClassAnalysis documentation built on July 12, 2019, 12:53 a.m.