Intelligent Sensing Analysis Using Mel-Time-Frequency-Imaging and Deep Learning for Distributed Fiber-Optic Vibration Detection

In fiber-optic vibration-sensing applications, intelligent sensing analysis is an important task for reducing costs and improving the overall quality of monitoring. Most existing methods show a low accuracy because their feature extractions significantly relied on the prior knowledge and the ability of the classifier designer. To deal with the challenges, an effective and scalable feature engineering method that uses vibration time-frequency imaging and deep learning has been proposed in this study. In order to improve the effectiveness and scalability, a Mel time-frequency-based vibration imaging approach is developed. This transforms the original time-domain data into a high-dimensional feature space and incorporates more feature information from data with various scales. Moreover, since the vibration-sensing recognition network is designed based on the deep-learning model, more discriminative features from the vibration image can be extracted automatically. The whole sensing recognition procedure includes four key steps: 1) original signal preprocessing of segmentation and denoising; 2) Mel time-frequency image generation; 3) representative feature extraction; and 4) vibration pattern-sensing classifier design. To verify the effectiveness and superiority of the proposed approach, eight datasets are collected, and three comparison schemes are investigated, using a distributed optical fiber interferometer-based vibration-sensing system. Our results demonstrate that the proposed intelligent sensing detection approach can provide higher accuracy and efficiency than other reported schemes.