Estimating Economic Activity by Applying Computer Vision To Satellite Imagery
Many underdeveloped countries lack proper funding to regularly collect granular data on economic activities. This makes it hard for these countries to address issues pertaining to divergences in regional growth, allocation of public resources, and sound investments of national money. The emergence of widely-available remotely sensed imagery in conjunction with machine learning, can help create or complement official statistics reflecting subnational economic activity.
The goal of my research is to apply computer vision – a branch of machine learning techniques – against satellite imagery in order to estimate economic activity in developing countries. As a first step towards this goal, I am working with imagery and measurements of economic activity from two states in the USA, Massachusetts and Pennsylvania. These data are more easily available, which is necessary to prove the concept of my research. More specifically, I aim to try and understand how different satellite imagery sources – which vary in temporal frequency, spatial resolution, and the availability of different spectral bands – affect the predictive capabilities of models for estimating economic activity. With this information, I intend to highlight the relative advantages and drawbacks of using different imagery datasets for estimating economic activity.
Research Area | Presenter | Title | Keywords |
---|---|---|---|
Artificial Intelligence | Rottenberg, Jacob | Computer Vision |