sits (version 1.1.0)

sits_uncertainty_sampling: Suggest samples for enhancing classification accuracy

Description

Suggest samples for regions of high uncertainty as predicted by the model. The function selects data points that have confused an algorithm. These points don't have labels and need be manually labelled by experts and then used to increase the classification's training set.

This function is best used in the following context

• 1. Select an initial set of samples.

• 2. Train a machine learning model.

• 3. Build a data cube and classify it using the model.

• 4. Run a Bayesian smoothing in the resulting probability cube.

• 5. Create an uncertainty cube.

• 6. Perform uncertainty sampling.

The Bayesian smoothing procedure will reduce the classification outliers and thus increase the likelihood that the resulting pixels with high uncertainty have meaningful information.

Usage

sits_uncertainty_sampling(
  uncert_cube,
  n = 100,
  min_uncert = 0.4,
  sampling_window = 10
)

Value

A tibble with longitude and latitude in WGS84 with locations which have high uncertainty and meet the minimum distance criteria.

Arguments

uncert_cube

An uncertainty cube. See sits_uncertainty.

n

Number of suggested points.

min_uncert

Minimum uncertainty value to select a sample.

sampling_window

Window size for collecting points (in pixels). The minimum window size is 10.

Author

Alber Sanchez,

Rolf Simoes,

Felipe Carvalho,

Gilberto Camara,

References

Robert Monarch, "Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI". Manning Publications, 2021.

Examples

if (sits_run_examples()) {
    # create a data cube
    data_dir <- system.file("extdata/raster/mod13q1", package = "sits")
    cube <- sits_cube(
        source = "BDC",
        collection = "MOD13Q1-6",
        data_dir = data_dir,
        delim = "_",
        parse_info = c("X1", "X2", "tile", "band", "date")
    )
    # build a random forest model
    samples_ndvi <- sits_select(samples_modis_4bands, bands = c("NDVI"))
    rfor_model <- sits_train(samples_ndvi, ml_method = sits_rfor())
    # classify the cube
    probs_cube <- sits_classify(data = cube, ml_model = rfor_model)
    # create an uncertainty cube
    uncert_cube <- sits_uncertainty(probs_cube)
    # obtain a new set of samples for active learning
    # the samples are located in uncertain places
    new_samples <- sits_uncertainty_sampling(uncert_cube)
}