GeoAgent is an open-source framework designed to facilitate Earth Observation (EO) tasks by leveraging state-of-the-art deep learning and machine learning tools. GeoAgent provides an accessible and scalable solution for researchers and practitioners working with EO data, integrating diverse data sources, models, and libraries to support tasks ranging from land cover classification to high-resolution image segmentation.
GeoAgent integrates with various open-source EO data sources in Google Earth Engine (GEE), enabling the retrieval and processing of data from the Sentinel-1 and Sentinel-2 satellites, Dynamic Land Cover, and more. This setup allows GeoAgent to handle open-domain EO tasks with ease.
GeoAgent incorporates popular deep learning libraries like terratorch and torchgeo, providing built-in support for datasets and models that can be trained with default configurations. These libraries also support domain-specific geospatial foundation models from IBM, including the NASA-IBM GFM Prithvi models. Currently, the framework includes Sentinel-1 and Sentinel-2-based deep learning models for key tasks: Multitemporal Crop Classification Burn Scar Segmentation Forest Degradation Segmentation Flood Mapping
For local studies requiring high-resolution RGB images, GeoAgent includes universal object detection and segmentation tools like samgeo and efficient-yolo-world. These tools enhance GeoAgent's capacity to perform detailed image analysis and segmentation for a variety of applications.
Although deep learning models form the core of GeoAgent, traditional machine learning tools are included as well, adding flexibility for tasks where simpler or interpretable models are beneficial.
To get started with GeoAgent, clone the repository and follow the installation steps for the necessary dependencies. Refer to the examples folder for detailed use cases and sample code.
#TODO
#TODO pip install -r requirements.txt Prerequisites Python 3.8+ Google Earth Engine API access Deep learning libraries: terratorch, torchgeo Object detection and segmentation tools: samgeo, efficient-yolo-world
Once installed, GeoAgent provides an interface to access and process EO data, #TODO.
Contributions are welcome! Please see our contributing guidelines for details on how to contribute to this project.
GeoAgent is licensed under the MIT License. See LICENSE for more information.
Thanks for the help from (https://github.com/NirantK/agentai)