Cocoa Segmentation in Satellite Images with deep learning


Ee0b4d48 a177 4b57 843d e0e97d444df6.png 500 500

Starting Date: earliest start: (2017-11-01) latest end: (2018-12-31)

Organization: Photogrammetry and Remote Sensing (Prof. Schindler)

Involved Host(s): Berger Monique

Abstract: Cocoa is the basic ingredient to produce chocolate.For market research, big chocolate companies need to know the overall acre-age of cocoa and a rough estimation of annual yield to adjust their buying strategy accordingly.

Description: Cocoa (see figures) is the basic ingredient to produce chocolate. It is grown in many tropical countries like Ivory Coast, Ghana, and Ecuador. For market research, big chocolate companies need to know the overall acreage of cocoa and a rough estimation of annual yield to adjust their buying strategy accordingly. Today, this is done manually by sending experts into the field that regularly measure the size of cacao beans and estimate the area of cocoa plantations. The aim of this project is to combine satellite images of the new ESA constellation Sentinel-2 and deep learning to segment cocoa planting sites in Africa and Latin America. What makes this task hard is the high similarity of cocoa and surrounding plants, often smallholder farms especially in Africa, and the inhomogeneous acquisition frequency due to frequent cloud coverage.

Goal: The candidate will develop a deep learning method and implement a software to segment cocoa at large scale.

Contact Details: Dr. Jan Dirk Wegner (jan.wegner@geod.baug.ethz.ch)

Deep learning satellite remote sensing semantic segmentation large-scale


Labels: Bachelor Thesis Master Thesis CLS Student Project (MPG ETH CLS)
Topics: Information, Computing and Communication Sciences Engineering and Technology

© 2017, Copyright Max-Planck-Gesellschaft - Imprint