Author: Wang, Biwei
Title: Application of neural networks and image processing techniques in distributed optical fiber sensor systems
Advisors: Yu, Changyuan (EIE)
Degree: Ph.D.
Year: 2023
Subject: Optical fiber detectors
Fiber optics
Hong Kong Polytechnic University -- Dissertations
Department: Department of Electronic and Information Engineering
Pages: xxiv, 158 pages : color illustrations
Language: English
Abstract: Distributed optical fiber sensor (DOFS) techniques have attracted a lot of interest in both research and application areas due to their capabilities of measuring temperature, strain, and vibration at consecutive points over a very long distance. Among all DOFS techniques, Brillouin optical time domain analyzer (BOTDA) and phase-sensitive optical time-domain reflectometry (Φ-OTDR) are two of the most popular directions in recent years. Inspired by the successful application of advanced machine learning and image processing techniques in the field of optical fiber communication, while the DOFS systems are very similar to optical fiber communication systems in fact, this thesis focuses on the application of some neural network (NN) techniques, including artificial neural network (ANN), convolutional neural network (CNN), and deep neural networks (DNN), and also an image processing method, the video block-matching and 3D filtering (VBM3D) in BOTDA and Φ-OTDR to improve the system performance and solve existing problems.
First, the DNN-based temperature distribution extraction method for using the BOTDA system is demonstrated experimentally. After appropriate training of DNN model, temperature distribution information along the fiber under test (FUT) could be directly extracted from the experimentally obtained local Brillouin gain spectrums (BGS) using DNN without the need of calculating Brillouin frequency shift (BFS) and transforming it to temperature as the conventional Lorentz curve fitting (LCF) method does. The results of temperature extraction using DNN show a comparable accuracy to that of using conventional LCF method, which proves that DNN can be used in BOTDA for temperature extraction.
Second, simultaneous temperature and strain measurement by using DNN for BOTDA has been demonstrated with enhanced accuracy. After trained by using combined ideal clean and noisy BGSs, the DNN is applied to extract both the temperature and strain directly from the measured double-peak BGS in large-effective-area fiber (LEAF). Both simulated and experimental data under different temperature and strain conditions have been used to verify the reliability of DNN based simultaneous temperature and strain measurement, and to demonstrate its advantages over BOTDA with the conventional equation-solving method. Avoiding the small matrix determinant induced large error, the DNN approach significantly improves the measurement accuracy. The enhanced accuracy and fast processing speed make the DNN approach a practical way of achieving simultaneous temperature and strain measurement by the conventional BOTDA system without adding extra system complexity.
In addition, a method of robust and fast temperature extraction for BOTDA using the denoising autoencoder (DAE) based DNN is demonstrated. After appropriate training, the DAE suppresses the noise on the measured BGS, and improves the signal-to-noise ratio (SNR) by 9.96dB in the experiment. To extract temperature, the DAE as a basic block is stacked to form the DNN model. Since the DNN model is based on DAE, both denoising and fast temperature extraction can be simultaneously finished using only one DNN model. Moreover, since the temperature information can be extracted directly from the experimental BGS data, the speed of temperature extraction using the DAE based DNN is faster by 500 times than that using LCF. Combining the advantages of both denoising and fast processing speed, the DAE based DNN would be a practical way of temperature extraction for the BOTDA systems.
Besides, the video-BM3D denoising method is proposed and experimentally demonstrated for the first time in a 100.8km long-distance BOTDA sensing system with 2m spatial resolution. Both experiments under static and slowly varying temperature environment are carried out. A temperature uncertainty of 0.43°C has been achieved with denoising by VBM3D in static temperature measurement. The slowly varying temperature at the end of 100.8km fiber has also been accurately measured. VBM3D exploits both the spatial and temporal correlations of the data for denoising, thus it can significantly reduce the temperature fluctuations and keep the measured values close to the real temperature even if the temperature is temporally changing. Thus, it would be useful for the long-distance sensing where the measurand may have temporal evolution in the slowly varying environment.
Finally, the ANN and CNN are applied in the Φ-OTDR, which is an optic fiber distributed acoustic sensing (DAS) system, to detect the sound and to predict the existence of red palm weevil (RPW). RPW is a detrimental pest, which has wiped out many palm tree farms worldwide. However, early detection of RPW is challenging, especially in large-scale farms. Here, machine learning techniques including the ANN and CNN are combined with the DAS as a solution for the early detection of RPW.
Within the laboratory environment, we reconstructed the conditions of a farm that includes an infested tree with ~12 day old weevil larvae and another healthy tree. Meanwhile, some noise sources are introduced, including wind and bird sounds around the trees. After training with the experimental time- and frequency-domain data provided by the DAS system, a fully-connected ANN and a CNN can efficiently recognize the healthy and infested trees with high classification accuracy values (99.9% by ANN with temporal data and 99.7% by CNN with spectral data, in reasonable noise conditions). This work paves the way for deploying the high efficiency and cost-effective fiber optic DAS to monitor RPW in open-air and large-scale farms containing thousands of trees.
Rights: All rights reserved
Access: open access

Files in This Item:
File Description SizeFormat 
6920.pdfFor All Users3.81 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: