Practical multi-class event classification approach for distributed vibration sensing using deep dual path network.

Area of Science:

• Engineering
• Signal Processing
• Machine Learning

Background:

• Multi-class event classification in distributed vibration sensing (DVS) is challenging due to ambient noise and nonstationary signals.
• Applications like perimeter security, railway safety, and pipeline surveillance require robust DVS performance.

Purpose of the Study:

• To develop a reliable and robust deep learning approach for multi-class event classification in DVS.
• To leverage spatial-temporal-frequency spectrum datasets for improved signal analysis.

Main Methods:

• A deep dual path network with high learning capacity was employed.
• Spatial time-frequency spectrum datasets were constructed using multidimensional DVS signal information, emphasizing spatial domain data.
• A high-parameter-efficiency network was utilized.

Main Results:

• The proposed scheme demonstrated good reliability and robustness in classifying DVS signals.
• Field tests in a railway safety monitoring scenario validated the approach.
• Seven types of real-life disturbances were accurately classified with f1-scores reaching 97%.

Conclusions:

• The developed deep dual path network effectively addresses the challenges of multi-class event classification in DVS.
• The novel spatial time-frequency spectrum datasets and network architecture significantly improve performance.
• This approach offers a viable solution for enhancing DVS performance in practical applications.