Using Deep Learning to Detect Vehicle Related Signals From Seismic Records
Description:
In the interest of safety and security, it is imperative for research and industry facilities to effectively monitor vehicle movements within their premises. While cameras have traditionally been employed for this purpose, concerns over privacy and susceptibility to light or weather conditions have prompted the exploration of alternative technologies. Seismic sensors, due to their unobtrusive nature and ability to capture ground motions generated by vehicle movements, present a promising solution. To assess the feasibility of utilizing seismic sensors for vehicle detection, we conducted a comprehensive study. We deployed three sets of nodal-type seismic stations in proximity to a road at the main campus of Oak Ridge National Laboratory. Each station set was collocated with a camera to cross-reference and validate the data. The labels of detected vehicles were processed using the images recorded by the cameras. We then use these labels, along with the corresponding seismic signals, to train a deep learning model. Over the course of the station deployment, an extensive dataset comprising more than 500,000 seismic signals and labels was gathered. Using this dataset, we trained a deep learning model designed to categorize whether a seismic signal contains a vehicle-related signature or not. As the deep learning model can use either a seismogram or a spectrogram as input, we evaluated the model performance for both cases. The results show that using a spectrogram leads to better model performance. Our model achieved an impressive F1-score exceeding 0.9. These findings underscore the potential of seismic sensors as a valuable complementary tool for vehicle monitoring, offering a resilient and privacy-conscious alternative to camera-based systems.
Session: ESC-SSA Joint Session: Climate Change and Environmental Seismology [Poster Session]
Type: Poster
Date: 5/3/2024
Presentation Time: 08:00 AM (local time)
Presenting Author: Chengping
Student Presenter: No
Invited Presentation:
Authors
Chengping Chai Presenting Author Corresponding Author chaic@ornl.gov Oak Ridge National Laboratory |
Omar Marcillo marcillooe@ornl.gov Oak Ridge National Laboratory |
Monica Maceira maceiram@ornl.gov Oak Ridge National Laboratory |
Ryan Kerekes kerekesra@ornl.gov Oak Ridge National Laboratory |
|
|
|
|
|
Using Deep Learning to Detect Vehicle Related Signals From Seismic Records
Category
ESC-SSA Joint Session: Climate Change and Environmental Seismology