Classify time series or data cube using machine learning models

Description

This function classifies a set of time series or data cube given a trained model prediction model created by sits_train.

SITS supports the following models:

• support vector machines: see sits_svm

• random forests: see sits_rfor

• linear discriminant analysis: see sits_lda

• quadratic discriminant analysis: see sits_qda

• multinomial logit: see sits_mlr

• extreme gradient boosting: see sits_xgboost

• multi-layer perceptrons: see sits_mlp

• mixed 1D CNN and MLP networks: see sits_TempCNN

• deep residual netwroks:see sits_ResNet

Usage

sits_classify(data, ml_model, ...)

## S3 method for class 'sits'
sits_classify(data, ml_model, ..., filter_fn = NULL, multicores = 2)

## S3 method for class 'raster_cube'
sits_classify(
  data,
  ml_model,
  ...,
  roi = NULL,
  filter_fn = NULL,
  impute_fn = sits_impute_linear(),
  interp_fn = NULL,
  compose_fn = NULL,
  start_date = NULL,
  end_date = NULL,
  memsize = 8,
  multicores = 2,
  output_dir = tempdir(),
  version = "v1",
  verbose = FALSE,
  progress = FALSE
)

Arguments

Value

Predicted data (classified time series) or a data cube with probabilities for each class.

Note

The "roi" parameter defines a region of interest. It can be an sf_object, a shapefile, or a bounding box vector with named XY values ("xmin", "xmax", "ymin", "ymax") or named lat/long values ("lat_min", "lat_max", "lon_min", "lon_max")

The "filter_fn" parameter specifies a smoothing filter to be applied to time series for reducing noise. Currently, options include Savtizky-Golay (see sits_sgolay) and Whittaker (see sits_whittaker).

The "impute_fn" function is used to remove invalid or cloudy pixels from time series. The default is a linear interpolator, available in sits_impute_linear. Users can add their custom functions.

The "interp_fn" function is used when the training samples which were used to generate the classification model have a larger number of time instances than the data cube. In this case, pixel time series of the data cube will have to be interpolated to fit that of the samples.

The "compose_fn" function is used when the training samples which were used to generate the classification model have a small number of time instances than the data cube. In this case, pixel time series of the data cube will have to be composed and/or merged to fit that of the samples.

The "memsize" and "multicores" parameters are used for multiprocessing. The "multicores" parameter defines the number of cores used for processing. The "memsize" parameter controls the amount of memory available for classification.

Author(s)

Rolf Simoes, rolf.simoes@inpe.br

Gilberto Camara, gilberto.camara@inpe.br

Examples

# Example of classification of a time series
# Retrieve the samples for Mato Grosso
# select an extreme gradient boosting model
samples_2bands <- sits_select(samples_modis_4bands,
                            bands = c("NDVI", "EVI"))
xgb_model <- sits_train(samples_2bands,
    ml_method = sits_xgboost(verbose = FALSE)
)
# classify the point
point_2bands <- sits_select(samples_mt_6bands,
                            bands = c("NDVI", "EVI"))
point_class <- sits_classify(point_2bands, xgb_model)
plot(point_class)

# create a data cube based on files
data_dir <- system.file("extdata/raster/mod13q1", package = "sits")
cube <- sits_cube(
    source = "LOCAL",
    name = "sinop-2014",
    origin = "BDC",
    collection = "MOD13Q1-6",
    data_dir = data_dir,
    delim = "_",
    parse_info = c("X1", "X2", "tile", "band", "date")
)

# classify the raster image
probs_cube <- sits_classify(cube,
    ml_model = xgb_model,
    output_dir = tempdir(),
    memsize = 4, multicores = 2
)

# label the classified image
label_cube <- sits_label_classification(probs_cube, output_dir = tempdir())
plot(label_cube)