Research On Signal Recognition Technology In Optical Fiber Security System

As people pay more attention to their personal and property safety in recent years,Optical fiber vibration sensing perimeter alarm system has been widely used in security for large-scale defense area such as large storehouse,oil pipelines,large hydropower equipment which cannot be taken care of by human resources and monitoring equipment.The main indicator for evaluating the performance of a perimeter security system is the ability to identify signal types correctly in real time.Currently,signal classification based on pattern recognition is the mainstream practice for classifying fiber sensing signals.The current fiber Optic Vibration Sensing Perimeter Security System mainly has the following problems:Firstly,as an expert system,the prior knowledge implanted to systems is particularly important,in this system it means the methods of extracting feature.At present,the feature extraction of most optical fiber vibration sensing perimeter alarm system is mainly based on frequency domain and wavelet domain feature extraction,the extracted content mainly includes some energy characteristics in the frequency domain and failure to fully exploit the inherent characteristics of the perimeter signal,therefore,there is a situation in which the accuracy of recognition of some burst signals is extremely low.There is no theoretical determination of the rationality of the extracted feature values,often resulting in redundancy between certain features.Secondly,as a classification system,the selection of classification algorithms is also crucial.In addition,as a system that needs to interact with the user and adapt to different environments,the system parameters should be adjusted automatically under different environmental conditions instead of making long-term manual adjustments every time the environment changes.Aim to solve the three main problems mentioned above,the main contents of this system are as follows:(1)Based on the characteristics of short-time energy and short-time zero-crossing rate in the time domain of optical fiber vibration sensing signals,a double threshold method is designed to filter the signal and filter out the signals with obvious time domain characteristics.Transform the signal to the frequency domain by Fast Fourier transform,Further screening of signals by frequency cutoff,Filter out signals with obvious frequency domain characteristics.(2)The voice signal is non-stationary,discontinuous,short-term stationary signal in long-term,which is similar to the fiber vibration signal.We learn from the method of voice signal processing to process the signal in the time domain,frequency domain and wavelet domain,and analyze the difference between different environmental signals and intrusion signals in each domain,and then extract the feature values on each domain of the signal as the classification algorithm.Enter the characteristics.The redundancy and rationality of feature extraction are analyzed according to the feature fusion theory.(3)We use two types of classification algorithms,support vector machine and probabilistic neural network,to classify signals.(4)We designed a self-adjustment strategy for system parameters based on Q learning to reduce manual adjustment of the system,which enhances the automation performance while improving the recognition accuracy of the system.