Crop segmentation from multispectral imagery
Orion Lab has assembled a novel benchmark dataset, namely Sen4AgriNet [1], for crop classification and segmentation via multi-temporal multispectral satellite imagery. The dataset comprises multi-temporal image acquisitions from the Sentinel-2 satellites over Catalonia and France, and the ground truth labels cover 168 classes.
The aim of this thesis is to experiment with different Deep Learning models for the segmentation of crops, and/or the detection of parcels.
Prerequisites: Strong Python skills, Machine Learning basic concepts, Deep Learning python framework (Pytorch, Tensorflow, etc)
Supervisor: Maria Sdraka