Classifying the raw irregular time series (CRIT) codes using Transformer for the paper: Zhang, H. K., Luo, D., & Li, Z. (2024). Classifying raw irregular time series (CRIT) for large area land cover mapping by adapting transformer model. Science of Remote Sensing, 100123. https://doi.org/10.1016/j.srs.2024.100123
Pro_load_model_run_tile_v2_5.py >> Use the trained CRIT model for land cover classification, including code to load ARD tiles, DEM, and the trained models for land cover mapping.
model.zip has the trained model and mean/std files.
The DEM data defined in Collection-2 Landsat ARD tiles can be find in https://zenodo.org/records/13875964.
The reference training and testing data saved in csv can be found in https://zenodo.org/records/8097697 (LCMAP_CU_Landsat_ARD.DAILY.85.06.18.24997.sensor.st.dem.csv)
Pro_lcmap_CRIT_v11_69.py >> CRIT codes including loading csv file, pre-processing data, splitting train and test data, CRIT training and evaluation, and saving trained model split folder: the split file use to define training and testing split that generate the results in Zhang et al. (2024)
Other files used in comparison experiments in Zhang et al. (2024): v11_694 1D-CNN to classify daily raw irregular time series with DEM/xy v9_31 CRIT without DEM/xy and v9_34 1D CNN for daily raw irregular time series without DEM/xy v9_21 Transformer for 16-day composites and v9_24 1D CNN for 16-day composites v9_52 Transformer for percentiles and v9_53 1D CNN for percentiles