Research On The Identification Method Of Intrusion Events In The Fiber-optic Zone Perimeter

With the rapid development of fiber-optic sensing technology,its application in the field of security has gradually shown its distinctive advantages.However,in the fiber-optic perimeter security system,man-made intrusion events are often misjudged as natural intrusion events,which not only trifles away a mass of human power and physical resources,but also seriously endangers people's life and property.Therefore,how to design a reliable and efficient intrusion detection system to quickly and correctly distinguish man-made intrusion events and natural intrusion events,which has attracted the attention of many researchers.But so far,this is still a difficult problem to be solved.Aiming at how to reduce the nuisance alarm rate of the invasive event,this paper analyzes the three aspects of the choice of optical path structure,the optimization of signal feature extraction algorithm and the selection of neural network model.Therefore,two different intrusion event recognition methods are presented.The main research contents are as follows:1.This paper introduces the single-mode—multimode—single-mode(SMS)fiber structure as the sensor in the zone perimeter,and collects the four typical intrusion events,namely man-made intrusion events(knocking,rattling)and natural intrusion events(wind and rain).2.For the intrusion signal generated from the multimode fiber,two different feature extraction methods are proposed.One is to use wavelet threshold denoising and short-time Fourier transform(STFT)to perform time-frequency analysis on the intrusion signal to obtain spectrogram,and then convert the spectrogram to black and white image and divide it into training set and test set;The second is to perform filter bank on the intrusion signal,and extract the features using singular value and kurtosis as combined vectors in different frequency channels after filtering,and divide the obtained feature vectors into training set and test set.3.According to the different feature expression forms of the intrusion signal,two different neural network models are selected for identification and analysis.One is to design a convolutional neural network(CNN)model with black and white image as the input form;the other is to design a probabilistic neural network(PNN)model with the feature vector as the input form,and employs the salp swarm algorithm(SSA)to optimize the smoothing factor of PNN model.Like most metaheuristic algorithms based on swarm mechanism,SSA has the disadvantages of poor convergence and being easily trapped in a local optimum.Therefore,we embed the weight factor and the adaptive mutation operator into the SSA to better optimize the smoothing factor of the PNN model.4.We have deeply discussed and analyzed the two different intrusion event identification methods mentioned above.A large number of experimental results show that compared with the CNN model,the PNN model optimized by the improved salp swarm algorithm(ISSA)can more efficiently identify and distinguish between man-made intrusion signals and natural intrusion signals,thereby reducing the nuisance alarm rate of intrusion signal and increasing the application value of SMS fiber structure in the zone perimeter.