Register now After registration you will be able to apply for this opportunity online.
AI-Powered Clustering of Satellite Data for Global Navigation
Imagine a robot that can intelligently interpret the world from above—looking at its immediate surroundings and predict how the terrain extends beyond its sensors. If the ground beneath its feet is muddy, it might infer that the meadows ahead will have a similar challenge, adjusting its path accordingly. To enable this kind of intelligent decision-making, robots need to recognize how local conditions propagate globally.
Machine learning techniques, particularly deep learning models, offer a way to automate this understanding by clustering satellite images based on visual and structural similarities.
Keywords: Learning, Deep Learning, Satellite Data, Global Planning, Robotics
The goal of this project is to develop and train a model that can cluster satellite images into meaningful groups based on visual similarity, geographic features, or correlated patterns of change. This includes both differentiating visually similar areas, but also correlating visually different clusters. This clustering will help identify areas in a global map that share common characteristics, which will help global navigation as one of the major downstream tasks. The project will involve training a neural network to extract features from satellite imagery and use these features for unsupervised clustering analysis.
The dataset will be sourced from Swisstopo (Federal Office of Topography), which provides high-resolution satellite data and aerial imagery of Switzerland across multiple temporal periods.
[1] Gómez-Chova, L., Tuia, D., Moser, G. and Camps-Valls, G., 2015. Multimodal classification of remote sensing images: A review and future directions. Proceedings of the IEEE, 103(9), pp.1560-1584.
[2] https://ml-gis-service.com/index.php/2020/10/14/data-science-unsupervised-classification-of-satellite-images-with-k-means-algorithm/
The goal of this project is to develop and train a model that can cluster satellite images into meaningful groups based on visual similarity, geographic features, or correlated patterns of change. This includes both differentiating visually similar areas, but also correlating visually different clusters. This clustering will help identify areas in a global map that share common characteristics, which will help global navigation as one of the major downstream tasks. The project will involve training a neural network to extract features from satellite imagery and use these features for unsupervised clustering analysis. The dataset will be sourced from Swisstopo (Federal Office of Topography), which provides high-resolution satellite data and aerial imagery of Switzerland across multiple temporal periods.
[1] Gómez-Chova, L., Tuia, D., Moser, G. and Camps-Valls, G., 2015. Multimodal classification of remote sensing images: A review and future directions. Proceedings of the IEEE, 103(9), pp.1560-1584.