基于小波包变换的光纤扰动信号模式识别

Pattern Recognition of Fiber Disturbance Signals Based on Wavelet Packet Transformation

神经网络是一种普遍采用的模式识别方法。当样本数目较大时,神经网络会因输入特征向量矩阵的庞大使结构变得复杂,运行缓慢,性能下降。针对这一问题,本文对光纤扰动信号提出了基于小波包变换的特征提取分类方法,以小波包系数的能量值作为信号的特征向量,用少量的特征向量反映信号的大部分特征信息,减少神经网络的输入和运行时间的同时保证识别的准确率。将现场采集的各类光纤扰动信号经由神经网络进行判别,实验表明,基于小波包变换的特征提取能在较短的时间内以较高的识别准确率分类出各光纤扰动信号的类别。

Neural network is generally used for pattern recognition. With larger number of samples, neural network will have more complex structure, slower running time and lower performance. In order to solve this problem, this paper proposes feature extraction and classification method based on wavelet packet decomposition to deal with fiber disturbance signals, which elects energy value of wavelet packet coefficients as signal feature vector, extracts appropriate feature vector to reflect fundamental characteristics of signals and reduces input data of neural network and running time while guaranteeing accuracy rate of classification. By experimenting with fiber disturbance signals collected from real field, results shows that feature extraction method based on wavelet packet decomposition can classify different kinds of fiber disturbance signals from real field with short time and high accuracy rate.