This repository provides code and supplementary materials for the paper entitled 'Remote sensing framework for lithological mapping via stacked autoencoders and clustering'. We present a framework based on different dimensionality reduction methods, including principal component analysis, canonical autoencoders, stacked autoencoders, and the k-means clustering algorithm to generate clustered maps using multispectral remote sensing data which are interpreted as lithological maps.
- Linux and Windows are supported, but we recommend Linux for performance and compatibility reasons.
- 1+ high-end NVIDIA GPU for sampling and 1+ GPUs for training.
- 64-bit Python 3.9 and PyTorch 2.1 (or later). See https://pytorch.org for PyTorch install instructions.
- Other Python libraries:
pip install click Pillow psutil requests scipy tqdm diffusers==0.26.3 accelerate==0.27.2
This framework is applied to three different data types, including are in /datasets/main_dataset/*.zip folder or can be download from the source attached below.
A small dataset also included in the datasets/sample_dataset.zip folder.
-
Landsat-8 download here
-
ASTER download here
-
Sentinel-2 download here
-
We share the code for each dataset separately including all the experiments.
-
Note that the specific dataloader, data preprocessing and postprocessing should be done by users depending on particular datasets.
Autoencoder_Landsat8.ipynbAutoencoder_ASTER.ipynbAutoencoder_Sentinel2.ipynb
Copyright © 2024,Transitional Artificial Intelligence Research Group & AFFILIATES. All rights reserved.
All material, including source code and pre-trained models, is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
Nagar, S., Farahbakhsh, E., Awange, J., Chandra, R., Remote sensing framework for lithological mapping via stacked autoencoders and clustering [Under Review]