Multiple Event Recognition Scheme Using Variational Mode Decomposition-Based Hybrid Feature Extraction in Fiber Optic DAS System
Wenqiang Fu, Duo Yi, Zhewen Huang, Chengdao Huang, Youfu Geng, Xuejin Li
Abstract
This study proposes a multiple intrusion event recognition scheme in a fiber optic distributed acoustic sensing (DAS) system. The variational mode decomposition (VMD) algorithm is selected for signal decomposition, and VMD-based short-time energy encoding is first proposed for feature extraction. In addition, short-time energy encoding is combined with VMD-based short-time zero-crossing rate and kurtosis to form the hybrid feature vector, and therefore, sufficient feature information of the intrusion event signals can be well-retrieved, which eventually improves the recognition performances. The experimental results demonstrate that the no-human wind disturbance event and three types of human intrusion events can be 100% successfully recognized, and the overall recognition rate and recognition time of four types of intrusion events are 98.4% and 0.207 s, respectively, by employing the optimized recognition scheme, which has been improved over the previously reported event recognition schemes. Furthermore, we preliminarily demonstrate the multipoint and multievent recognition based on the proposed fiber optic DAS system with an optimized event recognition scheme, showing great potential for intelligent security monitoring applications.