Machine learning projects require data – that goes without saying. However, obtaining high-quality data for training and testing purposes is one of the most pressing challenges facing AI teams today. For example, Cognlytica found that some 25% of the time spent on machine learning projects was spent on data sourcing…
A U.S. company was building a model to detect changing environmental landscapes and they needed to segment large satellite images so that their model could identify different types of land use.
Aya Data’s challenge was to segment roads, rivers, lakes, residential use, industrial use and natural terrain into different classes within 2,000 ultra-high resolution satellite images of South America.
Identifying human-led deforestation was a particular challenge within this project.
Aya Data used its image segmentation experts in Ghana to accurately label 2,000 ultra-high resolution satellite images. Teams experienced in labelling satellite imagery were critical to delivering the project to the right degree of accuracy within the given timeframe.
Computer Vision Vs Environmental Degradation
The complexity and scale of each image meant that the images had to be viewed in part and as a whole to provide perspective and context to larger-scale objects such as man-made rivers originating at hydro-electrical plants.
The video images labeled by Aya Data were used to accurately train a custom computer vision model to detect the changing land uses in regions of South America.