Abstract: Information extraction and recognition methods of optical fiber vibration signals have gradually become the focus of research. Visual analysis and comparison of the short-time zero crossing rate and energy of 5 kinds of optical fiber vibration signals, such as excavator digging, artificial digging, automobile walking, pedestrian walking and noise, are carried out, and a method for selecting experimental samples is proposed. A decision tree model of two classification and ELM algorithm are adopted, and according to the importance of the events, they are identified at four stages. At the same time, the influence of the parameters about ELM algorithm on the experimental results is analyzed. It is proved by experiments that the recognition rate of events is improved and the training time of model is shortened.

Key words: event, optical fiber vibration signal, experimental sample, Extreme Learning Machine（ELM）, recognition rate

摘要： 光纤振动信号的信息提取与识别方法逐渐成为研究热点。对挖掘机挖掘、人工挖掘、汽车行走、人员行走和噪声这五种光纤振动信号的短时过零率和能量特征进行可视化分析，提出一种实验样本的选取方法；采用二分类任务决策树模型和ELM算法，根据事件的重要程度分四个阶段完成事件的识别。探讨ELM算法中各参数对实验结果的影响。通过实验证明，该方法提高了事件的正确识别率，大大缩短了模型训练时间。

关键词: 事件, 光纤振动信号, 实验样本, 极限学习机（ELM）, 识别率