Creates sample data for Convolutional Neural Network

This function creates an array of input images and associated responses which can be utilized to train a Convolutional Neural Network.

Usage

cnn_make_samples(
  data,
  x,
  y,
  response,
  raster,
  raster_padding = FALSE,
  padding_method = NULL,
  size = 5
)

Arguments

data

data.frame or tibble. Database that includes longitude, latitude, and response columns.

x

string. Specifying the name of the column with longitude data.

y

string. Specifying the name of the column with latitude data.

response

string. Specifying the name of the column with response.

raster

SpatRaster. Raster from which predictor data will be cropped.

raster_padding

logical. If TRUE, the raster will be padded when cropping extends beyond its boundaries. Useful for ensuring all focal cells have the same size output even at the edges of the raster. Default FALSE

padding_method

string or NULL. Method used for padding the raster if raster_padding is TRUE. Options are "mean", "median", "zero". Ignored if raster_padding is FALSE. Default NULL

size

numeric. Size of the cropped raster, number o cell in each direction of a focal cell

Value

A list with two elements - 'predict' (a list of input images) and 'response' (an of response values). Each element in the 'predictors' list is an array representing a cropped image from the input raster.

Examples

if (FALSE) {
require(dplyr)
require(terra)

# Load data
envar <- system.file("external/envar.tif", package = "adm") %>%
  rast()
data("sppabund")
some_sp <- sppabund %>%
  filter(species == "Species one")

cnn_samples <- cnn_make_samples(
  data = some_sp,
  x = "x", # x coordinates for each point
  y = "y", # y coordinates for each point
  response = "ind_ha",
  raster = envar[[c("bio12", "sand", "elevation")]],
  size = 5 # how many pixels from point to border?
)

length(cnn_samples$predictors) #  938 matrix sets
dim(cnn_samples$predictors[[1]]) # three 11x11 channels
cnn_samples$predictors[[1]] # representing predictor variables
rast(cnn_samples$predictors[[1]]) %>% plot()

cnn_samples$response[[1]] # linked to a label
}