"Monitoring earthquakes is very important, it helps us better understand the cracks and tremors under vulnerable cities. If we know how the cracks develop, then we can predict aftershocks," says a researcher at Stanford University.Gregory Barrosa.
As mentioned earlier, the noise of cars, planes, helicopters, and the general city noise makes it difficult to detect the underground signals that indicate that an earthquake is either already occurring or will occur.
To make the process of detecting earthquakes easier, Barroso and his team created a neural network that was trained on about 80,000 city noise recordings and 33,751 earthquake signals. By combining them and eliminating noise, the artificial intelligence learned to filter out the sounds of the city.
"We created about a million different combinations of these sounds, all of which were studied by a neural network," says Barroso.
Recordings of noise and earthquakes were collected in Long Beach and San Jacinto, California. After training, when the audio was fed through a neural network, the researchers found that the background noise was reduced by 15 decibels, which can be considered a success for artificial intelligence.