Explainable AI for Geo

In this project, we developed an explainable AI workflow for detecting invasive and exotic tree species in the Atlantic Forest of Brazil using Unmanned Aerial Vehicle (UAV) images. The project was successfully completed through a collaboration between the eScience Center and the Faculty of ITC at the University of Twente.

During the development process, we utilized image data from UAVs/drones, which have been increasingly employed in developing countries for various applications. Compared to satellite imagery, UAV imagery captures more details and thus provides more locally relevant information. Leveraging this feature, we used UAV images together with human-labeled species as training data. Ultimately, a Deep Learning (DL) workflow was developed to detect the target species.

We encountered several challenges during the development of the DL workflow. First, Deep Learning methods are notorious for requiring extensive labeled datasets to train and evaluate their performance, which were limited in this project. Additionally, the lack of labeled data posed challenges for training models for new applications, such as detecting new species. To address the first challenge, we designed a workflow based on few-shot learning algorithms to overcome the limited availability of labeled data. Furthermore, by integrating a Siamese network with explainable AI (XAI), the workflow enables the classification of new tree species with minimal labeled data while providing visual, case-based explanations for its predictions. Results have demonstrated the effectiveness of the developed workflow in identifying new tree species, even under data-scarce conditions.