Signal Recognition of -OTDR System Based on Wavelet Packet Decomposition and SVM
Bu Zehua,Mao Bangning,Si Zhaopeng,Gong Huaping,Xu Ben,Kang Juan,Li Yi,Zhao Chunliu
DOI: https://doi.org/10.3788/gzxb20225111.1106003
IF: 0.6
2022-01-01
ACTA PHOTONICA SINICA
Abstract:With the rapid development of the Internet of Things era, the perception and acquisition of all kinds of external information is becoming increasingly important. Traditional monitoring and warning technologies such as manual inspection, video surveillance, infrared detection, and microwave detection have been widely used in various fields. Although these technologies have succeeded, they are expensive, environmentally sensitive, and cannot be monitored over long distances. To realize real-time and environment-free long-distance monitoring, distributed optical fiber sensing technology has developed rapidly in recent years. Distributed optical fiber sensing uses the whole optical fiber as the signal's transmission medium and sensing unit. When external factors act on the optical fiber, the light wave transmitted in it is modulated accordingly, and its light intensity, frequency, phase or polarization state, and other parameters will change accordingly. Distributed optical fiber sensing has the advantages of flexible layout, high cost performance and wide measurement range. Compared with traditional methods, distributed optical fiber sensing realizes large-scale measurement at a lower cost. As an important branch of distributed optical fiber sensor, phase sensitive optical-time domain reflectometer has the advantages of high sensitivity, high resolution and simple structure. At present, phi-OTDR plays an important role in many applications, such as structural crack detection, railway monitoring, traffic flow detection, vehicle detection and intrusion detection. With the development of modern productive forces and the diversification of life scenes, simple vibration location and signal demodulation cannot meet the actual needs and user needs. Random interference in nature (such as thunderstorm, wind, hail, etc.), passing vehicles and moving personnel will interfere with the accurate recognition of intrusion signals, increasing the false alarm rate. It is hoped that the type of vibration signal can be obtained at the same time as the location of the vibration signal, so as to replace manual inspection more intelligently. Compared with traditional monitoring methods, phase sensitive optical-time domain reflectometer system has greater advantages in real-time performance and convenience. At the same time, compared with video surveillance, phase sensitive optical-time domain reflectometer system is more covert, has strong resistance to external electromagnetic and other interference signals, and has the advantages of low cost, wide range and continuity. Therefore, it is of great significance to classify and identify vibration signals, identify the types of vibration signals, and study related identification algorithms to improve the identification accuracy and response speed of vibration signals, which are of great significance to the monitoring occasions in the fields of road traffic and border security. In view of the phase-sensitive optical time domain reflectometer distributed optical fiber sensing system has difficulties in real-time performance and accuracy of signal recognition. A method based on wavelet packet decomposition and support vector machine is presented. The energy feature vector is extracted by the wavelet packet decomposition of the signal as the input samples of the support vector machine, and the energy distribution trend of different signals is analyzed. A total of 800 experimental samples of knock, shake, walk and noise signals were trained. The recognition effect was evaluated by four evaluation indexes:accuracy rate, recall rate, F1 value and accuracy. The experimental results show that the precision rate, recall rate and F1 value of the knocking signal are 94.12%, 96% and 95.05%, respectively;the precision rate, recall rate and F1 value of the shaking signal are 95.92%, 94% and 94.95%, respectively;The precision rate, recall rate and F1 value of walking signal and noise signal are all 100%. The overall recognition accuracy is above 97%. The method improves the recognition result accuracy and real-time performance in the signal recognition of phase sensitive optical-time domain reflectometer system.