Communications of the ACM
Skoltech Method Helps Train Computer Vision Algorithms on Limited Data
Researchers at the Skolkovo Institute of Science and Technology have found a way to help computer vision algorithms process satellite images of the Earth more accurately, even with very limited data for training.
Credit: Jose Luis Stephen/Getty Images
Researchers at Russia's Skolkovo Institute of Science and Technology (Skoltech) have developed a technique to help computer vision algorithms more accurately process satellite imagery of the Earth, even when trained on limited data.
Skoltech's Ivan Oseledets and colleagues formulated the MixChannel image augmentation method for multispectral satellite images, founded on substituting bands from original images with bands from images of the same area at a different time.
The team trained a convolutional neural network on six Sentinel-2 satellite images of forests in Russia's Arkhangelsk region.
Despite having just six images to train on, the MixChannel technique outperformed state-of-the-art solutions when tested with three neural networks, and it can be integrated with other augmentation methods for additional training data.
From Skolkovo Institute of Science and Technology (Russia) (07/15/21)
View Full Article
No entries found