Research On Pattern Recognition Based On Distributed Optical Fiber Vibration Sensor System

Railway is the main artery of national economy,and the importance of safety protection is self-evident.At present,the illegal intrusion along the railway mainly depends on active equipment and manual detection,which has a large amount of maintenance work,poor prevention and real-time performance,and high risk factor.The application of distributed optical fiber sensing technology(dofvs)as the safety monitoring along the railway has the characteristics of long monitoring distance,anti electromagnetic interference,almost "zero" cost maintenance work,and has a strong competitive advantage.Because of the high sensitivity of this sensing technology,the environment along the railway is complex and the background noise is serious,there are some false alarms for the perceived events that affect the safety of train operation.With the rapid development of modern artificial intelligence technology,such as big data analysis and application,machine learning and so on,it can greatly improve the recognition ability of abnormal events,reduce the false alarm rate and avoid missing alarm,and provide technical means for the safe real-time on-line monitoring of railway.In this paper,combined with the characteristics of the common abnormal events along the railway,using digital signal processing and artificial intelligence technology,the method of classification,judgment and recognition of the data sensed by DOFVS is proposed.The specific work includes:(1)According to the characteristics of optical fiber laying along the railway,the experimental site and optical fiber laying method are designed.The common intrusion events along the railway are simulated,and the signal acquisition is completed,including four kinds of vibration events and noise signals generated by personnel walking,driving,shovel digging and excavator operation.(2)An orthogonal matching pursuit based on sparse decomposition is proposed,(OMP)algorithm de-noising,establish the standard mode of signal de-noising,and implement the software code.Compared with the results of orthogonal decomposition wavelet threshold de-noising algorithm and set empirical mode decomposition method,the experiment shows that the signal-to-noise ratio is greatly improved,the root mean square error is almost zero,the smoothness is low,and the de-noising effect of dofvs sensing signal is obvious.(3)The time-frequency analysis of the denoised signal is carried out.Through the preprocessing of data normalization and framing,the convergence speed and effectivesignal purity of the algorithm are accelerated.The autocorrelation feature is improved,the mean correlation coefficient is used to judge and classify,and the feature extraction in time domain and frequency domain is realized.According to the orthogonality and difference of the multi feature joint,the feature reorganization is carried out,and the optimal multi-dimensional feature combination method is proposed.The results show that the recognition rate of vehicle driving events and mechanical mining events is improved by 9.09% and 6.25% respectively.(4)This paper proposes an event recognition method based on limit learning machine,which can improve the recognition accuracy of five common vibration events along the railway.Compared with the recognition results based on the random forest method,it can effectively reduce the false alarm and the false alarm.Using the confusion matrix to evaluate the accuracy,accuracy and sensitivity of the model,it is proved that the classifier based on limit learning machine has 100% accuracy and95.89% precision,recall 0.00%.In this paper,the method of event recognition based on OMP denoising,multi-dimensional feature combination and limit learning machine classification has better advantages,which provides technical support for real-time monitoring of illegal intrusion behavior and perimeter protection along the railway.