Optical Fiber Distributed Vibration Sensing Using Grayscale Image and Multi-Class Deep Learning Framework for Multi-Event Recognition

作者：Zhenshi Sun, Kun Liu, Junfeng Jiang, Tianhua Xu, Shuang Wang, Hairuo Guo, Zichun Zhou, Kang Xue, Yuelang Huang, Tiegen Liu · 发表于：IEEE Sensors Journal · 年份：2021 · DOI：10.1109/jsen.2021.3089004 · 被引用次数：49 · 研究领域：Advanced Fiber Optic Sensors、Structural Health Monitoring Techniques、Optical Coherence Tomography Applications

Multi-class sensing recognition is important in optical fiber distributed vibration systems, since accurate detections on the vibrations can significantly improve the performance of the systems in engineering applications. In this work, we have developed an ameliorated deep learning (DL) approach based on the serial fusion feature extraction model for multi-category sensing recognition in optical fiber distributed vibration sensing systems. Thus, the effective features can be extracted automatically and recognized accurately with the proposed approach. Firstly, we use a conversion algorithm to convert time-domain signals into the two-dimensional grayscale images. Then the proposed multi-class DL approach can be trained, in an end-to-end manner, to extract the convolutional features within each frame and the temporal features between frames. Finally, the extracted features are passed to a Softmax classifier to compute the final results. This proposed model is investigated to recognize seven common sensing patterns collected by a fiber-optic interferometer based vibration system in the field trial. Recognition accuracy and efficiency of the proposed DL approach have been greatly improved. Specifically, an average accuracy of 96.16% is achieved in the ten-fold-cross evaluation. The recognition response time per sample has been greatly shortened to 11 ms. The results have been compared to the performance of a conventional DL recognition model and a handcrafted feature extraction ...