摘要
针对现有人体行为识别系统识别精度不高,且不便于日常使用的缺点,提出了一种用于智能手表的神经网络分类算法。采用基于PCA的特征提取方法对Apple Watch智能手表采集到的三轴加速度数据进行特征提取,结合动量-自适应学习率BP神经网络分类算法有效识别出了行走、慢跑、上楼梯、下楼梯四种行为,识别准确率达到82.36%。与朴素贝叶斯算法和决策树分类算法进行对比实验,结果显示基于PCA的神经网络分类算法进行人体行为识别准确率更高。
Most of the existing human behavior recognition systems aren't convenient for daily use andcan't recognize different behavior effectively. In this paper, a neural network classification algorithm isproposed based on smart watch. A feature extraction method based on principal component analysis(PCA) is used to extract features from three axis acceleration data collected from Apple Watch. A BPneural network classification algorithm with momentum and adaptive learning-rate is used to recognizewalking, jogging, upstairs and downstairs effectively, and the recognition accuracy reaches 82.36%.Compared with the Naive Bayes and Decision tree algorithms, the experimental results show that theneural network algorithm based on PCA has better recognition accuracy.
出处
《电子设计工程》
2017年第4期27-31,共5页
Electronic Design Engineering
基金
河南省重点科技攻关项目(152102210249)
中国教育科研网下一代互联网技术创新项目(NGII20150704)