A comparative analysis involving dncnn and pca on ф-otdr vibration sensing

In Distributed Optical Fiber Sensors (DOFS), the Phase-sensitive Optical Time Domain Reflectometry (Ф-OTDR) technology has tremendously demonstrated stupefying performance with regard to measurements of real-time accurate positioning of trains, intrusion detection, all due to its unique prospects on high sensitivity and precision, fast speed response, long distance sensing, everlasting lifetime service, and above all, low operational cost. Nonetheless, its application for vibration detection becomes stressful should the data is impeded by harsh external conditions. Hence to successfully enhance its smooth application, we investigated and executed a robust deep learning algorithm-Denoising Convolutional Neural Network (DnCNN) on Ф-OTDR sensing data for vibration detection. We utilized 60 exquisite layers comprising ReLU, 2Dconvolutional, batch normalization in order to improve the training speed and denoising performanceand finally a regression layer. The trained network (TrainedNet), was successfully performed after obtaining Digital Down Conversion (DDC) of the Ф-OTDR noisy data. The target of locating the vibration point was smoothly harnessed at a distance of approximately 50m and the proposed DnCNN technique was then evaluated against one state-of-the-art denoising algorithm and it outperformed it. The theoretical analysis and simulated demonstrations of the preceding locations under the sensing distance of 200m are hereby presented as proof of concept.