Detecting trees from satellite imagery

Problem Statement:

Trees are vital for carbon sequestration, urban heat mitigation, and public health, yet maintaining accurate inventories remains a challenge for many communities. This project addresses the critical need for automated tree canopy mapping by leveraging multiple satellite modalities. Students will develop a system that fuses optical and radar imagery to detect individual trees and quantify canopy cover across diverse landscapes. The resulting solution will empower evidence-based urban forestry, support carbon offset verification, and enable communities to monitor environmental justice and climate adaptation initiatives.

Task:

This assignment aims for:

• Development of an AI model that detects trees based on satellite imagery with high precision and recall;
• The AI-based system should be robust enough to work reliably across diverse environments and landscapes.
• The AI model developed should consider problems of occlusion and dense canopies, where the individual trees are impossible to be detected, and other algorithms need to be applied.

Work:

10% Theory, 60% Modelling, Coding and Testing, 10% Evaluation and Validation, 20% Writing

Contact:

Andreas Kamilaris: a.kamilaris@utwente.nl