DeepCrop🛰️🌾🌽

We are creating an AI system for nation-wide agricultural production monitoring based on crowd-sourced and remote-sensing.

My role

I am the Director of this project, which is a collaboration between the University of Chile, the Centre of Artificial Intelligence, the Center of Natural Ressources as principal institutions, and the Technical University of Munich, the European Commission’s Joint Research Center, and the École Polytechnique Fédérale de Lausanne.

This is a 4 years Tecnologia Avanzada project funded to the tune of 660,000,000 CLP¹ coming as grant from the National Research and Development Agency (ANID).

The project

This project aims to develop a crop map (like land-use but for crop i.e., which crop are cultivated where) at the country-level. To this aim we will leverage the capacity of general purpose model that we will trained over Chile. This is quite fun as Chile is a very long country with many different climates, making it the perfect place to test a model claiming to be general.

The objectives are:

• Gathering existing general crop data at the polygon- and pixel-level from open-source and in-house datasets
• Collecting Chilean crop data at the polygon- and pixel-level, including yield
• Implementing a parcel delineation model, with a polygon-level crop classifier
• Pre-training a large vision model on Worldwide, South American, and Chilean multimodal and multiresolution data
• Train the model to learn to learn various tasks using meta-learning algorithms
• Implement a dashboard using the model’s predictions