Counting Yurts in Mongolia Using Machine Learning

The project aimed to count all yurts in Mongolia using machine learning techniques without prior expertise in this area.

Yurts, also known as ger, are widely used in Mongolia, particularly in Ulaanbaatar, where many people live in ger districts due to rapid urbanization and housing shortages.

The project utilized satellite imagery and machine learning models for object detection to estimate the number of yurts in Mongolia, discovering 172,689 yurts with a high confidence level.

The challenges faced included the need for significant amounts of labeled data, optimizing the machine learning model's accuracy, and efficiently processing large amounts of geographical data.

An innovative approach was used to automate yurt identification by developing a backend model that improved through iterative feedback during the labeling process.

The study highlighted the socio-economic factors in Mongolia, including the transition from a nomadic society to urbanized areas, resulting in a mix of modern and traditional lifestyles.

Infrastructure development in Mongolia, particularly in urban areas like Ulaanbaatar, has not kept up with demand, leading to informal settlements in ger districts.

The study reflects on the broader implications of urbanization and industrialization in Mongolia and raises questions about policy effectiveness and housing challenges.

The research demonstrates the potential of machine learning and satellite imagery in socio-economic studies to provide unique insights.