This paper proposed a location independent multi-model mobile user behavior recognition method due to the fact that location diversity on mobile users influence the accuracy of their behavior recognition.After extracting the features of the mobile phone's acceleration signals,features' similarity could be calculated.Comparing this features similarity with the supposed data,in the meantime,setting the maximum features' similarity as testing data,the selected data would be classified by ELM (extreme learning machine).The experimental results show that this method can improve the recognition accuracy rate by 11％.Thus the experimental results is satisfying and this recognition method is effective.%针对智能手机佩戴位置多样性对移动用户行为识别结果的影响,提出一种位置无关的多模型移动用户行为识别方法.该方法通过计算手机加速度传感器所采集到的行为信号在不同佩戴位置的特征相似度,与预先计算的不同佩戴位置特征相似度进行比较,并采用相似度最大的位置特征作为测试样本,利用极速学习机(ex-treme learning machine,ELM)分类器对移动用户行为进行识别.实验结果证明,相对于不区分佩戴位置的行为识别方法,该方法可将识别准确率提高11％,是一种有效的移动用户行为识别方法.
展开▼
