Remote sensing can help to understand different aspects of our planet: land use, weather, human behavior, natural disasters and so on. Due to the massive data throughput of satellites, computer vision and machine learning algorithms are essential for taking full advantage of this source of knowledge. Otherwise, it would be impractical to produce results within an acceptable time frame. This line of research focuses on the automatic extraction of meaningful information from satellite imagery, while taking the computational cost into careful account.
Title: Resource-Constrained Geospatial Land Use Classification and Object Detection
Description: Land use classification and vehicle detection are critical pieces of information for a widerange of applications, from humanitarian to military purposes. Applying computer vision and machine learning algorithms to perform such tasks autonomously requires the ability to overcome different challenges, such as occlusions, varying perspectives, fine-grained discrimination, and class imbalance. This project explores ensembles of deep neural networks and test augmentation to tackle these problems.
R. Minetto, M. Pamplona Segundo, S. Sarkar (2019). "Hydra: An Ensemble of Convolutional Neural Networks for Geospatial Land Classification", IEEE Transactions on Geoscience and Remote Sensing.
G. Rotich, S. Aakur, R. Minetto, M. Pamplona Segundo, S. Sarkar (2018). "Using Semantic Relationships among Objects for Geospatial Land Use Classification", IEEE Applied Imagery Pattern Recognition Workshop.
G. Rotich, R. Minetto, S. Sarkar (2018). "Resource-Constrained Simultaneous Detection and Labeling of Objects in High-Resolution Satellite Images", arXiv.
Title: Economic Recovery Markers from Satellite Imagery to help with City-scale Decisions during COVID-19 Recovery
Description: Lockdowns and quarantines implemented worldwide due to the COVID-19 outbreak can be noticed even from space. Remote sensing can help to understand their effects on human and economic activities by recording changes in human behavior over time. This project focuses on the use of satellite imagery to support decisions concerning: (1) traffic issues, to ensure citizens' mobility but at the same time to avoid traffic jams that block the exchange of essential supplies; (2) facilities activity, to safely and economically maximize resources availability; and (3) social distancing, to appraise whether people are following health and safety directives issued by the government.
C. Hill (2021). "Automatic Detection of Vehicles in Satellite Images for Economic Monitoring", MS Thesis, University of South Florida.
R. Minetto, M. Pamplona Segundo, G. Rotich, S. Sarkar (2020). "Measuring Human and Economic Activity from Satellite Imagery to Support City-Scale Decision-Making during COVID-19 Pandemic", IEEE Transactions on Big Data.
Title: Airport traffic indicator based on flying airplane detection
Description: Air traffic is of particular interest to unveil human and economic activities (e.g., travel, tourism, cargo) and track disease spread due to in-flight transmission. This indicator is based on the detection of flying airplanes in images captured by the Copernicus Sentinel-2 satellites. To do so, we took advantage of the advances driven by deep learning algorithms − bio-inspired neural networks that learn representations with multiple abstraction levels and discover intricate patterns in massive data − to devise an automatic airplane detector.
M. Pamplona Segundo, A. Pinto, R. Minetto, R. Torres, S. Sarkar (2021). "Measuring Economic Activity From Space: A Case Study Using Flying Airplanes and COVID-19", IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.
M. Pamplona Segundo, R. Minetto, C. Hill, A. Pinto, R. Torres, S. Sarkar (2020). "Airport traffic indicator based on flying airplane detection", Rapid Action on coronavirus and EO, European Space Agency.