Projects | Martín Domínguez Durán

Reliable GEDI pairs for assessment of strctural changes in vegetation due to fires

Tue, 01 Jul 2025 00:00:00 +0000

For the first part of my PhD, I am working on improving the reliability of GEDI pairs for assessing structural changes in vegetation due to fires. The goal is to develop a reliable method that can be used to assess the impact of fires on vegetation structure using GEDI data, which is a spaceborne LiDAR mission that provides high-resolution measurements of forest structure.

HagueHaven: Hackathon for good challenge

Mon, 04 Dec 2023 00:00:00 +0000

How can we help municipality workers and housing organizations in The Hague (Municipality of The Hague) to allocate shelter seekers in better matching locations by enabling them to work efficiently within a shared overview?

This was the challenge that our team (Petra Bardócz, Annemieke Verhoek, Alejandro Salgueiro Dorado and I) decided to tackle during the 6th edition of Hackathon For Good.

Our solution focused on creating a centralized environment where all three parties (municipality social workers, housing organizations and shelter seeker) can participate in the process, take relevant actions, and update the offer and demand in a near real-time fashion. With our application we ended up winning the challenge and were also awarded the WorldStartup bonus prize for the most promising business opportunity with clear and realistic roadmap!

It was a great learning experience and an amazing 48 hour of hacking with an awesome team.

Bridging domains: Assessment of Domain Adaptation methods for mapping Cashew crops in Africa

Fri, 11 Aug 2023 00:00:00 +0000

Abstract

The recent implementation of commodity crop import regulations has highlighted the importance of refining crop mapping techniques from a regional to a global scale. Deep Learning models have shown great performance on this task. However, they struggle when trained on small or poor-quality datasets, and when implemented on a different domain than the one on which they were trained. Domain adaptation techniques emerge to address these issues. In this study, the performance of adversarial domain adaptation techniques was assessed for increasing the generalization capabilities of cashew crop semantic segmentation models trained on a source domain and applied to a target domain. Notably, an Attention U-Net + DANN model was trained using both a land cover benchmark dataset and a dataset created using PlanetScope images containing cashew crops from Ivory Coast (source domain) and Tanzania (target domain). Firstly, the shift between domains was evaluated through a time series analysis, an input and an inner feature distribution analysis. Secondly, the performance of the domain adaptation methods was assessed by comparing an unsupervised and a semi-supervised scenario with a lower-bound (source-only) and an upper-bound model (target-only). Finally, a web-application was built to locate cashew crops in the regions of interest and label data in these data-scarce regions. The proposed network reached a lower-bound F1-score of 0.02, an upper-bound F1-score of 0.62 and a relative improvement of 27% using domain adaptation. While the use of domain adaptation on the cashew dataset yielded a minor improvement, these methods demonstrated success on the semi-supervised scenario (92%) and on the benchmark dataset (45%). The low performances on the cashew dataset were attributed to the initial shift between the crop characteristics across the domains and the inherent difficulty faced by the domain adaptation method in aligning them. In addition, this study suggests that the domain shift observed is influenced by seasonal environmental variations, differences in cultivation practices, and potential sensor characteristic variations. Finally, the low performances of this study highlighted the need for improved data quality and expanded datasets in the domains studied.

Hear and Now: Interactive web application for bird diversity education

Wed, 07 Jun 2023 00:00:00 +0000

Hear and Now was born as a project for the Geo Information Science and Remote Sensing intergration course given for master’s students in Wageningen University & Research. During this course, together with other 5 master’s students, we visited the beautiful Achterhoek region in the Netherlands. During out time there we fell in love with the enchanting songs of birds in the region. Because of this, we developed a web application for bird watchers or bird waching enthusiasts like us. Our main objective with this application is to make people more aware of the biodivesity around them.

In our web application you will find what you need to boost your love for bird watching. In it you can either detect birds using audio files or check some birds that have been seen nearby. Right now the default coordinates are the ones in the Achterhoek region, nevertheless, you can simply change this by modifying the values in Latitude and Longitude.

Overview

Development of Indoor Rn modeling app

Sat, 04 Feb 2023 00:00:00 +0000

Radon (²²²Rn) is a naturally occurring gas that represents a health threat due to its causal relationship with lung cancer. Despite its potential health impacts, few studies have been carried out to assess this public health problem in countries where RC measurements and research are scarce. This project aims to contribute to the bridging of the baseline information gap by using inferential statistic methods to estimate indoor RC spatial distribution and building an easy-to-use webapp for RC modeling.

Overview