The Research Of Feature Extraction And Recognition Algorithm For Fiber Intrusion Signal On Small Sample

The optical fiber pre-warning system has been widely used in the field of pipeline protection for long-distance real-time monitoring due to its advantages of strong anti-interference ability and high accuracy.At present,the optical fiber sensors based on phase-sensitive optical time-domain reflectometer are commonly used.It senses external intrusion vibrations and realizes the detection and recognition of the optical fiber intrusion signals through various signal processing algorithms.Among the various types of vibration sources,the damage of mechanical intrusion is the highest and the waveform is regular.Therefore,there are many identification algorithms for mechanical signals with high recognition rate.In the actual environment,artificial intrusions occur more frequently with complex waveforms and similar frequencies,making it difficult to extract features.In this paper,the feature extraction of common artificial intrusion signals is researched,and the directional recognition of tapping signals and vehicle signals is realized.In order to identify multiple types of artificial signals at the same time,a variety of signal processing algorithms are studied and improved in this paper.The article first introduces several single parameter features,including over-average rate features,one order difference features to identify tapping signals and short-term average magnitude difference function features to identify vehicle signals.According to the characteristics of the signal,different feature extraction algorithms are designed,and we realize the directional recognition of single optical fiber artificial signal.Based on the characteristics of optical fiber artificial signals with obvious time domain differences and similar spectral distribution,a wavelet multi-resolution analysis method is proposed to simultaneously extract the feature of signals from both time domain and frequency domain.First,we perform wavelet multi-layer decomposition on the pre-processed fiber intrusion signal to obtain the wavelet coefficients of each frequency band.Then we calculate the energy ratio to obtain the feature vector.Finally,a random neural network RVFL is selected to train and test the feature vectors.This algorithm can effectively identify a variety of optical fiber intrusion signals and achieve a higher recognition rate.In addition,based on the characteristics of optical fiber signals with non-linear and non-stationary,this paper proposes a more adaptive feature extraction algorithm based on ensemble empirical mode decomposition.The algorithm can be decomposed according to the time scale of the signal itself,without setting any basis function and solving the mode mixing phenomenon caused by the acquisition of the optical fiber signal,and effectively recognizes three types of signals: tapping,vehicle and running.Based on this,we introduced the permutation entropy judgment mechanism to improve the algorithm.We choose decomposition algorithms with different complexity according to the randomness of the sequence,which reduces the intrinsic mode function that appear in the decomposition process.Finally,simulation experiments prove that the improved algorithm improves the recognition rate and reduces the time consumption of feature extraction.