Deep Learning Algorithms for Satellite Imagery

This book provides key insights into the world of Deep Learning pertaining to satellite image understanding. It highlights what differentiates satellite image datasets from other natural or synthetic images and how to tackle problems specific to these imagery data. From answering questions like how to select optimal training data to weekly supervised and unsupervised learning and how to tackle loosely labeled data, it is a valuable source of information for anyone interested in understanding the theory behind satellite image analytics and provides key insights on the application of various state-of-the-art Deep Learning algorithms on these datasets.

Unsupervised learning for satellite imagery. Working with loosely labeled data. Classification vs Segmentation. Semi-supervised learning. Using adversarial learning. Post-processing – Structured Prediction. Using hyperspectral images. Combining hand-crafted features and deep learning. Active Learning. Transfer Learning. Choosing optimal training data. Understanding relation between training data and optimal model size . Effect of data augmentation. Effects of skip connections for creating bigger networks. How satellite image processing differs from natural image processing.

Saikat Basu is working as a research scientist in the Facebook Maps team in Boston. He received his PhD in Computer Science from Louisiana State University in 2016. He received his Bachelor of Technology in Computer Science and Engineering from National Institute of Technology, Durgapur, India in 2011. During his doctoral program, he has been doing research on the analysis of various kinds of imagery data using Computer Vision and Deep Learning algorithms for the analysis of satellite imagery data. During his PhD, he has worked as a research associate at NASA Ames Research Center, Moffett Field, California and an intern at the Facebook Maps team in Boston.