Masters Thesis Presentation - Meerke Romeijnders

Title: Supporting Regreening Practices Through Image Classification and Geospatial Analysis

Abstract:

This work aims to address the gap in the practical application of regreening techniques among farmers using the JustDiggit educational app Kijani in Africa, where currently only about 20% of users translate digital learning into real-world action. To address that, we wanted to create an interactive feature to support the "treecovery" method, allowing users to take a picture of a tree they wish to prune and receive specific information about its type. This required developing a custom classification feature, as existing plant-identification applications are predominantly trained on data from Europe and North America, rendering them ineffective for local flora. The setup compared computer vision detection and classification models to identify the best performing solution for this specific task. By doing that, we were able to evaluate which model best bridges this technological divide. Ultimately, identifying the optimal model not only helps individual users apply these pruning techniques correctly but also supports the company in maximising its broader environmental and social impact.

Supervisors: Prof. Dimka Karastoyanova, Dr. Estefanía Talavera Martínez