sits_applysits_mixture_model for spectral
mixture analysissits_viewsits_as_sf to convert sits
objects to sfsits_regularizeroi parameter in sits_regularize
functioncrs parameter in sits_get_data"MPC"sits_whittaker() function to process
cube.sits_lighttae()
(Lightweight Temporal Self-Attention)sits_uncertainty_sampling() for active
learningsits_confidence_samples() for
semi-supervised learningsits_geo_dist() to generate samples-samples
and samples-predicted plotsits_tuning() for random search of machine
learning parameterssits_reduce_imbalance() function to balance
class samplessits_as_sf() to convert a sits tibble to a
sf objecttorchopt deep learning optimizer
packagesits_uncertainty(): least
confidence and margin of confidencesits_kfold_validate()data to samples in sits machine
learning classifiers (NOTE: models trained in previous versions is no
longer supported)file parameter in sits_get_data()
functiontorch
package and remove keras dependencesits_TAE() classification modelsits_lightgbm() classification modelsits_regularize() parameterssits_regularize() to reach production level
qualitysits_regularize() to use C++ internal
functionssits_cube() to open results cubeplot() parameters on raster cubessits_view()sits_get_data() to accept tibblessits_cube()sits_regularize() to process in parallel by
tiles, bands, and datessits_regularize() to check malformed filesAWS_NO_SIGN_REQUEST environment variable.gc_get_valid_interval() function.sits_regularize has a fault tolerance system, so
that if there is a processing error the function will delete the
malformed files and create them again.sits_regularize function has a new parameter called
multithreads.sits_cube function for local cubes has a
new parameter called multicores.F1 score in sits_kfold_validate with
more than 2 labels.sits_cube() function to tolerate malformed paths
from STAC service;sits_apply() function to generate new bands
from existing ones;sits_accuracy() function to work with multiple
cubes;sits_view()sits_uncertainty() function to provide
uncertainty measure to probability maps;sits_regularize() by taking least cloud cover
by default method to compose imagessits_regularize that generated images with
artifactssits_cube from STAC AWS
Sentinel-2sits_timeline() to sits model objectsconfig_colors.yml by removing palette
namessits_regularize()start_date and end_date from
validation csv filesits_regularize() is producing Float64 images
as outputgdalcubes_chunk_size in “config.yml” to improve
sits_regularize()..source_collection_access_test to pass
ellipsis to rstac::post_request function..source_collection_access_test to pass
ellipsis to rstac::post_request function.sits_plotsits_timeline for cubes that do not have the
same temporal extent.S2_10_16D_STK-1 removed from BDC source in
config fileNoClass label improvementmapview to leaflet packageCLASSIFIED and PROBS sources from
config fileterra package to
1.4-11sits_list_collections() to indicate open data
collectionptw,
signal and MASSopen_data collections in config
fileoutput_dir parametersits_cube_clone() functionsits_select() for bands in raster
cubesits_regularize()
functionOPENDATA sourceS2_10-1 BDC collection from configsits_list_collections().source_bands_resampling()sits_som_clean_samples() functionsits_bands<-() functionsits_select() functionsits_bbox() functionS2-SEN2COR_10_16D_STK-1 BDC collectioncheck functionsatellite and sensor info in config
fileimager, ranger, proto,
and future packages from sitssits_cube.local_cube() function parameters
satellite and sensororigin and collection to
sits_cube.local_cube() functionroi parameter in
sits_classify() functionRaster classification results can now have versions: a new
parameter “version” has been included in the sits_classify
function.
Corrections to sits_kohonen and to the
documentation.
New deep learning models for time series: 1D convolutional neural
networks (sits_FCN), combining 1D CNN and multi-layer
perceptron networks (sits_TempCNN), 1D version of ResNet
(sits_ResNet), and combination of long-short term memory
(LSTM) and 1D CNN (sits_LSTM_FCN).
New version of area accuracy measures that include Olofsson metrics ()
From version 0.8 onwards, the package has been designed to work with data cubes. All references to “coverage” have been replaced by references to “cubes”.
The classification of raster images using
sits_classify now produces images with the information on
the probability of each class for each pixel. This allows more
flexibility in the options for labeling the resulting probability raster
files.
The function sits_label_classification has been
introduced to generate a labelled image from the class probability
files, with optional smoothing. The choices are
smoothing = none (default),
smoothing = bayesian (for bayesian smoothing) and
smoothing = majority (for majority smoothing).
To better define a cube, the metadata tibble associated to a cube
requires four parameters to define the cube: (a) the web service that
provides time series or cubes; (b) the URL of the web service; (c) the
name of the satellite; (d) the name of the satellite sensor. If not
provided, these parameters are inferred for the sits
configuration file.
The functions that do data transformations, such as
sits_tasseled_cap and sits_savi now require a
sensor parameter (“MODIS” is the default)
Functions sits_bands and sits_labels
now work for both tibbles with time series and data cubes.
sits_show_config() to see the default contents. Users
can override these parameters or add their own by creating a
config.yml file in their home directory.Examples and demos that include classification of raster files
now use the inSitu R package, available using
devtools::install_github(e-sensing/inSitu).
All examples have been tested and checked for correctness.
sits_coverage has been replaced by
sits_cube.
sits_raster_classification has been removed. Please
use sits_classify.
In sits_classify, the parameter
out_prefix has been changed to output_dir, to
allow better control of the directory on which to write.
sits_bayes_smooth has been removed. Please use
sits_label_classification with
smoothing = bayesian.
To define a cube based on local files,
service = RASTER has been replaced by
service = LOCALHOST.
For programmers only: The sits_cube.R file now
includes many convenience functions to avoid using cumbersome indexes to
files and vector: .sits_raster_params,
.sits_cube_all_robjs, .sits_class_band_name,
.sits_cube_bands, .sits_cube_service,
.sits_cube_file, .sits_cube_files,
.sits_cube_labels, .sits_cube_timeline,
.sits_cube_robj, .sits_cube_all_robjs,
.sits_cube_missing_values,
.sits_cube_minimum_values,
.sits_cube_maximum_values,
.sits_cube_scale_factors, .sits_files_robj.
Please look at the documentation provided in the
sits_cube.R file.
For programmers only: The metadata that describes the data cube no longer stores the raster objects associated to the files associated with the cube.