Theoretical Basis Research On Detection And Recognition Method For ?OTDR Optical Fiber Intrusion

This paper focuses on the theoretical research of detection and recognition method for phase-sensitive optical time-domain reflectometer(?OOTDR)optical fiber intrusion.In order to achieve the harmful intrusion signal location and identification,a detection and recognition method is designed for optical fiber pre-warning system(OFPS)based on ?OTDR structure.In the first place,the classical constant false alarm rate(CFAR)methods are studied and a new CFAR method in spatial dimension is established in detection part.Non-parametric test is designed to eliminate harmless intrusions in time dimension.Then,the feature extraction methods are provided according to pitch period(PP),duty cycle(DC)and zero-crossing rate(ZCR)in recognition part.These features are basis of accurately identify harmful intrusions.Finally,the study of visual attention architecture(VAA)is carried out,so that the overall method structure is reasonable based on VAA.In view of this,a detection and recognition method is established based on VAA for OFPS.Moreover,the experiment is performed to verify effectivity of the method.In the first place,detection method for OFPS is studied.Background homogeneity adaptive(BHA-)CFAR is provided in this paper in spatial dimension.The detection method as much as possible reduces calculation load while ensuring the detection performance.It effectively eliminates noises.The non-parametric test methods,such as sequential probability rate test(SPRT)and Kolmogorov-Smirnov(K-S)test,are applied in time dimension.In other words,according to the frequent degree of harmless intrusions,SPRT and K-S test are used.It effectively removes harmless intrusion signals.Then,recognition method for OFPS is studied.Some phenomenon are observed that only mechanical intrusion signals have PP,just MD intrusion signals have relatively small DC,and nothing but PV intrusion signals have relative small ZCR.Moreover,it is verified by measured data experiments that different type harmful intrusions can extract the corresponding features.Finally,the design of whole detection and recognition method is inspired by VAA.The whole method consists of data-driven and task-driven two processes.Furthermore,it is verified by measured data experiments that later data quantity can be accurately reduced in data-driven process and recognition accuracy is improved in task-driven process.