Data Science and Remote Sensing
Scientific methods in Remote Sensing (RS) are changing because of the impact of information technology and data science advancements to extract knowledge or insights from data. The data deluge, triggered by ever-increasing data acquisition rates in combination with data curation improvements, leads to scientific methods using complex machines, which, on the one hand, requires access to comprehensive data sources and archives and, on the other hand, also supplements the linking via Linked Open Data (LOD) with a fundamental understanding of the information stored there by machines.
Automating quantitative information extraction from imagery has been a hot topic of research for quite a few decades. We have noticed a variety of unsupervised imagery classification algorithms get introduced into mainstream image processing and GIS software.
In future, with big data, an organization that has access to all imagery ever collected using satellites and all classification outputs (whether supervised or unsupervised) of those images, can feed this as input data as a massive training set to a deep learning algorithm. This training set would encompass images with various spatial and spectral resolutions, images from a variety of satellite sensors, images from various time periods and classification outputs with a variety of accuracy and output classes. Such a real-world training dataset is immensely valuable and we are likely to see a phenomenal advancement in automating information extraction.