sits_tae: Train a model using Temporal Self-Attention Encoder

In sits: Satellite Image Time Series Analysis for Earth Observation Data Cubes

Train a model using Temporal Self-Attention Encoder

Description

Implementation of Temporal Attention Encoder (TAE) for satellite image time series classification.

This function is based on the paper by Vivien Garnot referenced below and code available on github at https://github.com/VSainteuf/pytorch-psetae.

We also used the code made available by Maja Schneider in her work with Marco Körner referenced below and available at https://github.com/maja601/RC2020-psetae.

If you use this method, please cite Garnot's and Schneider's work.

Usage

sits_tae(
  samples = NULL,
  samples_validation = NULL,
  epochs = 150,
  batch_size = 64,
  validation_split = 0.2,
  optimizer = torchopt::optim_adamw,
  opt_hparams = list(lr = 0.001, eps = 1e-08, weight_decay = 1e-06),
  lr_decay_epochs = 1,
  lr_decay_rate = 0.95,
  patience = 20,
  min_delta = 0.01,
  verbose = FALSE
)

Arguments

samples	Time series with the training samples.
samples_validation	Time series with the validation samples. if the samples_validation parameter is provided, the validation_split parameter is ignored.
epochs	Number of iterations to train the model.
batch_size	Number of samples per gradient update.
validation_split	Number between 0 and 1. Fraction of training data to be used as validation data.
optimizer	Optimizer function to be used.
opt_hparams	Hyperparameters for optimizer: lr : Learning rate of the optimizer eps: Term added to the denominator to improve numerical stability. weight_decay: L2 regularization
lr_decay_epochs	Number of epochs to reduce learning rate.
lr_decay_rate	Decay factor for reducing learning rate.
patience	Number of epochs without improvements until training stops.
min_delta	Minimum improvement to reset the patience counter.
verbose	Verbosity mode (TRUE/FALSE). Default is FALSE.

Value

A fitted model to be used for classification.

Author(s)

Charlotte Pelletier, [e-mail]

Gilberto Camara, [e-mail]

Rolf Simoes, [e-mail]

References

Vivien Garnot, Loic Landrieu, Sebastien Giordano, and Nesrine Chehata, "Satellite Image Time Series Classification with Pixel-Set Encoders and Temporal Self-Attention", 2020 Conference on Computer Vision and Pattern Recognition. pages 12322-12331. DOI: 10.1109/CVPR42600.2020.01234

Schneider, Maja; Körner, Marco, "[Re] Satellite Image Time Series Classification with Pixel-Set Encoders and Temporal Self-Attention." ReScience C 7 (2), 2021. DOI: 10.5281/zenodo.4835356

Examples

if (sits_run_examples()) {
    # create a TAE model
    torch_model <- sits_train(samples_modis_ndvi, sits_tae())
    # plot the model
    plot(torch_model)
    # create a data cube from local files
    data_dir <- system.file("extdata/raster/mod13q1", package = "sits")
    cube <- sits_cube(
        source = "BDC",
        collection = "MOD13Q1-6",
        data_dir = data_dir
    )
    # classify a data cube
    probs_cube <- sits_classify(
        data = cube, ml_model = torch_model, output_dir = tempdir()
    )
    # plot the probability cube
    plot(probs_cube)
    # smooth the probability cube using Bayesian statistics
    bayes_cube <- sits_smooth(probs_cube, output_dir = tempdir())
    # plot the smoothed cube
    plot(bayes_cube)
    # label the probability cube
    label_cube <- sits_label_classification(
        bayes_cube,
        output_dir = tempdir()
    )
    # plot the labelled cube
    plot(label_cube)
}

sits documentation built on Nov. 2, 2023, 5:59 p.m.