Remote Sensing and Computer Vision Algorithms at Scale: Defense and Humanitarian Uses

Abstract

The objective of this thesis is to understand how cloud computing and artificial

intelligence can be applied to vast amounts of remotely sensed data to better understand

macro-level trends for humanitarian and defense issues. Computer vision algorithms and

data were provided by Orbital Insight, Inc., a geospatial analytics company based in Palo

Alto, CA.

Specific projects were curated, data was acquired, and analysis was applied to

three use cases: “Patterns of life for The Battle of Marawi”, “Indications and Warnings

using multi-class aircraft detections”, and “Camp Fire land cover analysis”. The use cases

show how with imagery ingestion pipelines, cloud computing, and computer vision

algorithms, a massive quantity of data can be analyzed in a relatively short amount of

time. Without these workflows and new technologies, analysis of large amounts of data

would prove to be less efficient and resource heavy. The events show the benefits users

of spatial data would have to gain a better understanding of humanitarian and defense

issues. Algorithms used to derive insights from thousands of imagery scenes consisted of

a car detection algorithm, multi-class aircraft algorithm, and a land cover classification

algorithm. Additionally, the thesis briefly explores the use of geolocation data to

supplement computer-vision algorithm data. The thesis shows on a high level, through

examples, how users could use the technologies to analyze data more efficiently. This

analysis can be incorporated into high-level humanitarian or defense decisions. Future

work regarding this field should seek to evaluate the algorithm performance on a more

granular level. Researchers should also build different algorithms on open-source

imagery to allow for more users to benefit from the efficiency computer vision provides.