Remote_sensing-forrests.

Unsupervised ML method is applied to create a classificator of forrests on LANDSAT images. Invariant Information Clustering algorithm is in use.

DATA:

Directory DATA/data_raw should contain a number of LANDSAT scenes as .tar archives. Example of the file is stored in the dir with only 1 band (channel). More raw data samples one can find by link https://drive.google.com/drive/folders/1NO5xrnibQDGEkF38lepmGcsank3Pqr-x?usp=sharing.

RawDataProcessing:

contains a code for making HDF5 file from raw images from DATA containig set of images in 7 spectral channels (datasets of shape (N,SIZE,SIZE,7)). Eventual HDF5 file is stored at './DATA/h5_files'.

TrainingNN

contains scripts for creating and training a neural network.

main.py:

creates and trains a ResNet neural network using TrainingNN scripts.

Some .py scripts have .ipynb clones.

How to launch

1. cropping_and_h5making.py
2. main.py
3. validation.py