Dataset "Authomatic landslide detection using deep learning approach"

Below, compressed in a rar file, we present the data set developed during the research "Authomatic landslides detection using deep learning approach", submitted to the journal Landslides.

“Landslide_database” folder: contains two ESRI shapefile-type vector files, the first (Ground_Truth_database), corresponds to the manually outlined landslides for training the deep learning model. The file contains the following fields: associated Sentinel 2 image for the bounding (S2), area, and perimeter. The second (Ground_Truth_database_centroid) corresponds to the centroid of the outlined landslides, in addition to the previously mentioned fields, it contains the X and Y coordinates.

“Model results” folder: contains an ESRI shapefile vector file (Pred_T18GYS), corresponding to the landslides detected and segmented by the deep learning algorithm in the Sentinel-2 test tile.

“Model validation” folder: contains multiple ESRI shapefile vector files used during model validation. Study area (Study_Area), roads (Roads) and populated areas (Localities). Contains the predicted landslides in the study area (Predict_T18GYS_SA), the randomly selected predicted landslides (Predict_T18GYS_Random) and their geometries (Predict_T18GYS_Geometry). The folder also contains manually outlined landslides (Groud_Truth_T18GYS) restricted to the extent of the evaluated tile, and outlined landslides that spatially match randomly selected predicted landslides (Ground_Truth_Random). Finally, the true positives (TP_T18GYS), false positives (FP_T18GYS) and false negatives (FN_T18GYS) are provided separately.