Kolkoo Landslides

prediction and ground truth together (green prediction, pink truth), bigger area
our prediction
correct img
prediction and ground truth together (yellow is prediction), very small image
evaluation of the bigger prediction (0s are landslides, 1s are not)

Inspiration

What it does

The main test_learn.ipynb expects a tif file, that represents the slopes of the area. It also expects the correct solutions of what areas are landslides.

Then, it selects a training area, which it converts to an array of 0s and 1s describing whether a given square is a landslide or not. Then it learns using linear model regression the relationship between the slopes in a square_size window with the landslide valuse. Finally, it produces shapefiles that can be visualized in QGIS which are our predictions of landslides in the designated test area.

There is a file compare.py, that evaluates our prediction based on black and white images.