Abstract: The use of mobile accelerometer in seismic scene recognition is a new hot spot in the field of earthquake early warning research. As to existing methods using smartphones for seismic scene recognition, their simulating seismic scene is far away from the real situation, and the recognition method is relatively rough. Using earthquake experiencing house to simulate seismic scene, a two-layer earthquake recognition method is proposed to distinguish seismic scene from human daily life scenes. Firstly, it selects an appropriate threshold for the proposed feature：Sum of Maximum Correlation（SMC）, excluding jogging, walking and other human activity scenes by data periodicity. Then in the second layer, it identifies seismic scene data from aperiodic data using [K] nearest neighbor algorithm. Experimental results show that the hierarchical recognition method is quick and recognition rate of both non-seismic and seismic scene are over 96%.

Key words: accelerometer, earthquake early warning, hierarchical recognition, single threshold, K-Nearest Neighbor（KNN）

摘要： 利用智能手机内置加速度计进行地震场景识别是地震预警研究领域的新热点，现有的利用手机识别地震的方法中，地震模拟场景与真实情形出入较大，且识别方法较为单一。利用地震体验屋模拟出地震场景数据，提出一种两层分类的地震场景识别方法以区分地震场景和不同日常生活场景。该方法中，首先为提出的最大相关系数和（Sum of Maximum Correlation，SMC）特征设置单门限，利用数据周期性排除跑步、步行等人类运动场景，再利用[K]最邻近算法对非周期性运动场景数据进行第二层分类，以正确提取地震场景。实验结果表明，该分层识别方法快速且地震场景和非地震场景识别率均在96%以上。

关键词: 加速度计, 地震预警, 分层识别, 单门限, [K]最邻近算法