Tree Identification on Satellite Images to Prevent Forest Fires
AI startup Spacept worked with 36 Omdena collaborators to build a deep learning model for tree identification. The model helps to prevent power outages and forest fires sparked by falling trees that are too close to power stations.
Deforestation, climate change, and the risk of wildfires are all directly linked. Growing deforestation impacts climate change, which increases the chances of the vegetation drying out, which in turn further increases the risk of fires.
According to Greenpeace, around 8 billion tons of CO2 are released by fire every year. This is about half as much as the emissions caused by the burning of coal around the world.
Mission-driven Swedish AI startup Spacept reached out to Omdena to build an AI solution to prevent forest fires. Their product fuses satellite images with machine learning to reduce the risk of power outages and fires sparked by falling trees and storms. The goals are to save lives and reduce CO2 emissions while also radically reducing time and cost for infrastructure inspection.
The video below shows how the community of data scientists, data engineers, and enthusiasts, came together to leverage various AI tools and methodologies to build the solutions.
The solutions: AI to prevent forest fires
Identifying trees on satellite images was a tricky task, and the collaborators combined human judgment with machine suggestions. The winning model, a deep U-Net model detected trees with 95% accuracy. This would not have been possible without the tremendous effort done by the community in terms of data labeling and data augmentation, e.g. by applying GANs. In addition, one task group built an elevation map to show forest cover.
Ultimately, these solutions are being implemented in Spacept’s product to help drive forward their mission of using AI to save lives, infrastructure costs, and to reduce CO2 emissions.
Impact beyond the project
A better way of detecting forest fires
In another Omdena project, our teams worked with Brazilian company Sintecsys to detect wildfires in close to real-time. The identification of trees from this project helped to facilitate the wildfire project. Currently, our machine learning models are being implemented in Sintecsys´s solution to prevent fires in the Amazon rain forest.
The self-organized system of several teams trying different approaches methods of solving the problem was great. It’s a win-win as the teams of engineers get the opportunity to leverage their skills on real-world problems and we find the best-fit solutions most efficiently.