Phase-OTDR Event Detection Using Image-Based Data Transformation and Deep Learning

TL;DR

The paper tackles Phase-OTDR event detection by converting 1D time-series traces into multi-channel RGB images using Gramian Angular Difference/Summation Fields and Recurrence Plots. This image-based representation enables leveraging transfer learning with pre-trained CNNs, achieving state-of-the-art-like accuracy for six event classes while drastically reducing data size. Through 5-fold cross-validation and holdout tests, the approach demonstrates robustness and strong generalization, with EfficientNetB0 and DenseNet121 reaching near 99% accuracy. The work provides public code and data, highlighting the practical potential for efficient, accurate fiber-optic monitoring in infrastructure and security applications.

Abstract

This study focuses on event detection in optical fibers, specifically classifying six events using the Phase-OTDR system. A novel approach is introduced to enhance Phase-OTDR data analysis by transforming 1D data into grayscale images through techniques such as Gramian Angular Difference Field, Gramian Angular Summation Field, and Recurrence Plot. These grayscale images are combined into a multi-channel RGB representation, enabling more robust and adaptable analysis using transfer learning models. The proposed methodology achieves high classification accuracies of 98.84% and 98.24% with the EfficientNetB0 and DenseNet121 models, respectively. A 5-fold cross-validation process confirms the reliability of these models, with test accuracy rates of 99.07% and 98.68%. Using a publicly available Phase-OTDR dataset, the study demonstrates an efficient approach to understanding optical fiber events while reducing dataset size and improving analysis efficiency. The results highlight the transformative potential of image-based analysis in interpreting complex fiber optic sensing data, offering significant advancements in the accuracy and reliability of fiber optic monitoring systems. The codes and the corresponding image-based dataset are made publicly available on GitHub to support further research: https://github.com/miralab-ai/Phase-OTDR-event-detection.

Figures (11)

• Figure 1: Transforming 1D Phase-OTDR data into images and employing deep learning models for event classification based on created dataset.
• Figure 2: The setup of the Phase-OTDR system for event detection (adapted from Cao et al.).
• Figure 3: Interpolated spatial-temporal samples showing distinctive characteristics of different events: (a) Background, (b) Digging, (c) Knocking, (d) Watering, (e) Shaking, and (f) Walking. Raw data without preprocessing demonstrates natural event characteristics with blue lines indicating 12 distinct fiber regions.
• Figure 4: Image-based data transformation workflow for Phase-OTDR event detection. (a) Raw 1D time series data from 12 spatial fiber regions, (b) Transformation into individual 500×500 pixel images using GADF, GASF, and RP techniques organized in 3×4 grid, (c) Final RGB images with downsampled 224×224 resolution for deep learning compatibility.
• Figure 5: Exploring signal transformations: GASF, GADF, and RP image encoding methods unveil varied behaviors across different sinusoidal signals. Signal 1: A pure sinusoidal signal with an amplitude of 4 and a frequency of 6 Hz. Signal 2: Another sinusoidal signal with an amplitude of 4 and a frequency of 3 Hz. Signal 3: Similar to Signal 1 but with added random noise, creating a sinusoidal signal with an amplitude of 4 and a frequency of 6 Hz.