Thesis subject
MSc thesis topic: Deep learning architectures for crop-type classification from remote sensing imagery
Crop type classification is a common task in agriculture, where remote sensing imagery is used to detect types of crops grown in fields. There are several machine vision architectures explored in this task with a varying degree of success, and recently a global benchmark dataset has been released (CropHarvest).
In this thesis you will develop several deep learning architectures for crop type classification from remote sensing images, and demonstrate their performance with a real dataset from Flevopolder, the Netherlands. The available data consists of 10 m Sentinel 2 data, with several images throughout the crop growing season.
The thesis will compare alternative architectures for the crop type classification task, and employ alternative architectures for addressing the problem (i.e. semantic classification vs. object detection). The temporal dimension in the series of images and the class imbalance in the crop labels are unique characteristics of the problem that need to be addressed.
Objectives
- Review the literature for deep learning architectures for crop type classification
- Implement alternative deep learning architectures for crop type classification
- Evaluate performance for the Flevenpolder dataset and a public benchmark
Literature
Requirements
Modelling & visualisation