Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Pengfei Wang
Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Kai Ding
Science and Technology on Near-Surface Detection Laboratory, Wuxi, China
Hao Li
Science and Technology on Near-Surface Detection Laboratory, Wuxi, China
Jianguo Zhang
Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Xin Liu
Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Qing Bai
Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Dong Wang
Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Key Laboratory of Advanced Transducers and Intelligent Control Systems (Ministry of Education and Shanxi Province), College of Physics and Optoelectronics, Taiyuan University of Technology, Taiyuan, China
Invasion incident pattern recognition is crucial for a distributed optical fiber vibration sensing system based on a phase-sensitive time-domain reflectometer. Despite traditional pattern recognition identifying the vibration signal, the classification accuracy needs to be improved and the classifier requires probabilistic output, in order to ameliorate the performance of pattern recognition. A novel pattern recognition method is proposed in this paper. The characteristic vector is extracted from the original vibration signal by wavelet energy spectrum analysis. The probabilistic output is realized by the classification algorithm of a relevance vector machine. The optimal decomposition layer of the wavelet energy spectrum analysis is determined as six layers because of the compromise between the classification accuracy and the computational complexity. Taking into consideration the ground material and the weather, the experiments of three vibration patterns are carried out including walking through the fiber, striking on the fiber, and jogging along the fiber at 2, 5, and 8 km of the sensing fiber. With the help of 10-fold cross validation, the multi-classification confusion matrix is obtained in order to clarify the correct and incorrect classification results. Moreover, the performance measures, involving precision, recall rate, f-measure, and accuracy, are then analyzed. A classification macro-accuracy of 88.60% is finally obtained.
