Classifying urban areas from learned acoustic and seismic data

Abstract

In this work, a deep transfer learning algorithm is used to classify different urban areas from acoustic and seismic data. The deep transfer learning model combines Google’s deep learning model AlexNet. Measurements of acoustic and seismic ambient noise were conducted in an urban environment. The urban acoustic and seismic measurements are heavily influenced by traffic noise but exhibit more variation with respect to urban location due to the influence of subsurface conditions. A K-means clustering analysis is employed on the acoustic and seismic spectrogram to classify the urban areas. The results demonstrate that seismic and acoustic data can have similar cluster centroid locations in the frequency band of 4 Hz–10 Hz. The transfer learning results will be presented and discussed.