Exploring aerial imagery for classifying land cover and usage
Datasets and description: https://csc.lsu.edu/~saikat/deepsat/
The labelled datasets consist of 28x28x4 images, with each image classified into land cover classes. Images were extracted from larger aerial image tiles (~6000 x ~7000 pixels), and selected to represent balanced and non-overlapping training sets.
Development of a process for classifying land cover from this training set will be further developed and adapted for use on satellite imagery.
Final presentation in PDF form is here