Image Segmentation for Dust Detection Using Semi-supervised Machine Learning

摘要: Dust plumes originating from the Earth’s major arid and semi-arid areas can significantly affect climate system human health. Many existing methods have been developed to identify dust non-dust pixels a remote sensing point of view. However, these use empirical rules therefore difficulty detecting above or below detectable thresholds. Supervised machine learning also applied detect satellite imagery, but are limited especially when applying outside training data due inadequate amount ground truth data. In this work, we proposed an automatic segmentation framework using semi-supervised learning, based on collocated dataset Visible Infrared Imaging Radiometer Suite (VIIRS) Cloud-Aerosol Lidar Pathfinder Satellite Observations (CALIPSO). The method utilizes unsupervised for VIIRS leverages guidance labels profile product CALIPSO determine clusters as final product. determined similarity spectral signature along tracks. Experiment results show that accuracy outperforms traditional physical infrared addition, performs consistently over three different study areas, North Atlantic Ocean, East Asia, Northern Africa.