We’ve all heard about illegal and uncontrolled deforestation around the globe. It is a severe problem for Amazon forests, Siberia and many other regions, and its drastic impact on environment cannot be underestimated. Most commonly, unauthorized tree cutting is carried out to clear the lands for agricultural needs or further timber trade. Quite often the measures taken to fight against these actions are inefficient, time-consuming and very costly, both in terms of investments and manual labor.
During this talk I will demonstrate how Machine Learning can be utilized to perform both more efficient and less expensive deforestation detection based on public satellite images. On top of that, we will have a look at how this solution was adjusted to solve the task of finding poorly used fields overgrown with trees and bushes.
We will go through the specifics of the data and the task itself, challenges we’ve encountered and how the system was put into production.