Artificial Intelligence can erase clouds from satellite images

The Artificial Intelligence powered by NVIDIA graphics cards help researchers eliminate clouds satellite images.

Researchers from the University of Osaka have shown that it is possible to remove clouds from satellite photographs thanks to artificial intelligence .

Through a study published in Science Direct , scientists from the university’s Division of Sustainable Energy and Environmental Engineering used an “adversarial generative network” or GAN with deep learning powered by NVIDIA graphics cards.

How did they manage to get clouds from satellite images?

“By training the generative network to ‘trick’ the discriminating network into thinking that an image is real, we get reconstructed images that are more self-consistent,” said first author Kazunosuke Ikeno in a statement.

GANs are based on a pair of networks that compete with each other to create realistic images. These competitive networks allow developers to train the artificial intelligences with less data .

The first, known as the “generative network”, reconstructs the images without clouds. The second, a “discriminative network”, uses a convolutional neural network to choose the images created by the first network and the actual photos.

The two competing networks enhance each other without the need for so much data: the result is very realistic images without clouds.

Using 400-by-400-pixel images, the researchers trained the models on a PC running the Ubuntu open-source operating system and a GeForce GTX 1060 GPU.

The method proposed in this study can be used in situations where buildings are hidden by thin clouds and are visible. It can also be used to automatically generate monitored data sets for objects such as roads and rivers in aerial photography, as well as buildings viewed from street level.