Research

I share a few interesting research papers on GeoAI:

• Bootstrapping Rare Object Detection in High-Resolution Satellite Imagery - Developed novel cluster-based sampling approaches that improve rare object detection efficiency from 2% to 30% without any initial labeled data. Applied to identify animal enclosures in Serengeti Mara region, enabling effective mapping with minimal labeling budgets.

• Core-Set Selection for Data-efficient Land Cover Segmentation - Introduced six new methods for selecting optimal training samples in remote sensing datasets. Demonstrated that training on strategically selected subsets can outperform using complete datasets, advancing data-centric approaches for Earth observation tasks.

• Sims: An Interactive Tool for Geospatial Matching and Clustering - Created a no-code web tool using Google Earth Engine that enables users to perform advanced clustering and similarity search over regions of interest. Designed for feature exploration rather than model creation, demonstrated through analysis of maize yield data in Rwanda.

• Sequence to sequence weather forecasting with long short-term memory recurrent neural networks - Applied multi-stacked LSTMs to predict temperature, humidity, and wind speed for 9 Moroccan cities. Using 15 years of hourly data, created models forecasting 24 and 72 hours ahead that outperform traditional methods.