Research On Optical Fiber Intrusion Signal Recognition Algorithms Based On Stochastic Learning Regularization

This paper mainly researches the fast and accurate optical fiber intrusion recognition algorithms for the Optical Fiber Pre-warning System(OFPS),which is based on phase sensitive optical time domain reflection technology(Φ-OTDR).Before model training,we need to perform data preprocessing on the optical fiber signals.Specifically,at the beginning,the detrending method is used to reduce the optical path jitter caused by the sensor drift phenomenon.then,in order to reduce the data distribution range and accelerate the convergence speed,a data normalization method is adopted to deal with it.finally,the Spectrum distribution characteristics of fiber signals are obtained by Fast Fourier transform(FFT)to prepare for the subsequent training of the fiber intrusion signal recognition model.And in terms of fiber signal classification models,for the traditional artificial neural network-based classifiers,which have long iteration times and high model complexity.In this paper,a randomiezd learning algorithm is introduced to accelerate the model's convergence speed while ensuring the learning ability of the recognition model,thereby reducing time consumption.However,in the actual training process,problems such as 'overfitting' and ill-conditions that occur in randomized learning algorithm models,which will cause model performance degradation and poor recognition accuracy.And for the sake of enhancing the generalization performance of the randomized learning algorithms and ensuring the stability of the recognition model,this paper conducts research on the optimization algorithms of the randomized learning algorithm model.It mainly includes L2 regularization,Truncated Singular Value Decomposition(TSVD)and Dropout optimization algorithms.In the research process,from algorithm principle to formula derivation,from objective function design to code implementation,through a comprehensive comparison and analysis of three regularization optimization methods,we show the performance improvement before and after model optimization.In the end,experimental results show that the optical fiber signal recognition methods based on the randomized learning algorithm model studied in this paper can effectively improve the recognition accuracy and maintain a low time delay.Besides,this research provides a feasible way to realize fast and accurate recognition of optical fiber intrusion signals in OFPS,and has certain research significance.