Welcome to the BiCubes project, supported by HFRI grant 3943. This project introduces a holistic and scalable approach to processing and analyzing Earth Observation (EO) data, focusing on data harmonization and machine learning analytics to enhance geospatial applications for monitoring land and water.
BiCubes aims to revolutionize the use of satellite imagery by creating Analysis Ready Data (ARD) and developing advanced machine learning frameworks. The project is divided into two main components:
-
Satellite Imagery Processing and Classification: This component focuses on processing and classifying satellite imagery using machine learning techniques, including transfer learning and Random Forest classifiers.
-
Analysis Ready Data (ARD) Production Pipeline: This component generates ARD from satellite imagery, ensuring the data is geometrically and radiometrically corrected, temporally consistent, and spatially enhanced.
- Data Harmonization: Develop methodologies to harmonize high-resolution, multitemporal EO data into cloud/shadow-free geospatial data cubes.
- Machine Learning Frameworks: Design robust machine learning frameworks for semantic information extraction and analytics, utilizing deep learning techniques such as generative adversarial networks and recurrent neural networks.
- Scalable Solutions: Exploit big data technologies and cloud environments to develop scalable solutions for monitoring land and water resources.
This repository includes scripts for:
- Image Classification: Using transfer learning with a pretrained ResNet50 model.
- Random Forest Classification: Training and evaluating classifiers for image segmentation.
- Temporal Feature Processing: Extracting and processing temporal features.
- Image Mosaicking: Creating mosaics from multiple images.
- Python packages:
numpy,tensorflow,keras,sklearn,osgeo,pandas,natsort,joblib,optparse,xlsxwriter,multiprocessing,rasterio,tqdm,argparse.
This repository provides a pipeline for:
- Virtual Dates Generation: Creating synthetic images for specified dates.
- Registration (AROP): Aligning images for geometric consistency.
- Sharpening: Enhancing spatial resolution using high-pass filter fusion.
- Cleanup: Organizing and cleaning up output data.
- Python packages:
GDAL,OpenCV,NumPy,Pandas,Rasterio,TQDM,Natsort,Openpyxl.
To set up the environment for both repositories, ensure you have Python 3.6 or higher and install the required packages using pip: bash pip install -r requirements.txt
Each repository contains detailed instructions on how to run the scripts and customize the processing for different datasets. Refer to the individual README.md files in each sub-repository for specific usage guidelines.
For any questions or issues, please contact the project team at karank@central.ntua.gr.
This project is supported by the BiCubes project (HFRI grant 3943). We acknowledge the contributions of all team members involved in the development and testing of these pipelines.
This project is licensed under the MIT License.