DL Algorithm Filter Out Urban Noises to Boost Earthquake Sensors

May 2, 2022
1 min read

More effective ways to detect an earthquake have been one of the most researched topics in disaster management of late. Previously, it was tough to discern tremors through underground signals being suppressed due to the hullaballoo of city traffic.

In order to distinguish the urban hustle and bustle and seismic signals, a group of researchers at Stanford University developed a deep learning algorithm. The neural network was trained and tested for validation across a combination of 80,000 samples of urban noise and 33,751 samples of earthquake signals. Both sample sources were located in California, and over a million combinations were tried to develop the most precise algorithm.

“Earthquake monitoring in urban settings is important because it helps us understand the fault systems that underlie vulnerable cities; by seeing where the faults go, we can better anticipate earthquake events,” stated Gregory Baroza of Stanford University, California.

Compared to existing de noisification practices, the neural network delineated earthquake signals by an average of 15 decibels, thrice of what was obtained in the former techniques.

However, a major shortcoming of the algorithm lies in the fact that it is trained via supervised learning. Here, humans labelled the datasets, which makes it quite a time-consuming approach.

In addition, for the application of this model in a wide field of effect, it needs to switch to unsupervised learning, where it gets to discover the ingrained structure of untagged datasets and therefore reduces human intervention to a large extent.

The group of researchers is also doubtful about how successful can the implementation of this algorithm will be across varied environments where noise signatures will invariably differ.