Characteristics Extraction And Recognition Research In Optical Fiber Pre-Warning Signal

This paper presents the feature extraction method of intrusion vibration signal in fiber optic early warning system.Firstly,the most basic single parameter features are introduced,including the duty cycle of the pick mattock signal and shovel signal,the pitch period characteristics of the electric drill and the electric pick signal,and the zero-crossing rate characteristics of vehicle passing signal.Herein,we need to find a method to analyze such optical fiber intrusion signals which are typically non-stationary signals.Therefore,the significance of cepstrum analysis of fiber vibration signal is proposed in this paper.Then,based on the principle of human auditory hearing,the method of MFCC feature extraction is given.In our study,we also find a new characteristic of mechanical signals which is its energy distributes on the high frequency band.The traditional frequency band division rules of MFCC will cause partially frequency information loss on the high frequency band for these mechanical signals leading to the decrease of recognition performance.In order to make full use of the spectrum information and extraction characteristics more effectively,this paper presents an improved MFCC algorithm for optical fiber vibration signals recognition.This method can effectively distinguish the three types of signals of electric pick,shovel and vehicle passing signal,and improve the recognition performance of electric pick signals at the same time.In addition,the exploration of the physical meaning involved in the intrusion signal is also essential.Conventional methods for analyzing the optical fiber intrusion signal are usually studied from a mathematical point of view on the one-dimensional signals.From a single dimension perspective,it is easy to lose the possible higher dimensional information,and more importantly,lack of exploration of the physical interpretation involved in the intrusion signal itself.In order to analyze the form of intrusion signal accurately under the physical location of specific intrusion source,we propose a method to extract the characteristic spectrum of the intrusion signal,which express the energy distribution of the intrusion signal in different frequency bands.In this paper,the time-frequency analysis is used to add the frequency dimensional information on the basis of spatial-temporal two-dimensional image representation of optical fiber intrusion signal.And the time-frequency analysis performed on the optical fiber intrusion signal is a more comprehensive method increasing the observed dimensions.Furthermore,the spatial-temporal-frequency 3-D imaging results are formed,which can be seen as a joint expression of the spatial-temporal distribution and the frequency characteristic of the fiber intrusion signal presented in the form of data cube stacked in frequency band sequence.On this basis,we can further extract the characteristic spectrums corresponding to each intrusion time and spatial position as the features of the optical fiber intrusion signals.This method may open up a new direction for further research on the feature extraction and type recognition of intrusion signals from a physical point of view.