Chameleon AI Program Speeds Up Satellite Image Classification

Researchers at EPFL, Wageningen University, MIT, Yale, and Jülich Research Center have developed METEOR, an AI program designed to classify objects in satellite images with minimal training data. Unlike traditional AI models that require extensive datasets, METEOR can recognize new objects after being shown only a few images, making it highly adaptable for environmental monitoring.

Key Findings

• METEOR uses model-agnostic meta-learning (MAML) to train on land cover classification tasks, allowing rapid adaptation to new image types

• The AI can identify ocean debris, deforestation zones, urban areas, and other environmental features with minimal labeled data

• Unlike conventional AI models, METEOR does not need retraining for each new object type, significantly reducing processing time

• The system was tested on five distinct tasks, including detecting changes in Beirut after the 2020 explosion and classifying urban land use

• This approach enhances satellite image analysis, improving efficiency in environmental science and disaster response

Citation

Tuia, D., et al. (2024). Chameleon AI program classifies objects in satellite images faster. Communications Earth & Environment.