面向人体行为识别的深度特征学习方法比较*

针对人体行为识别问题，比较了两种基于智能手机惯性加速度传感器数据的深度特征学习方法。与传统的人工特征提取方法相比，基于深度特征学习方法可以实现端到端训练，网络结构简单直观，避免了繁琐的特征工程，通过深度神经网络模型的学习自动获得特征。本文通过对比深度卷积神经网络、长短期记忆网络两种深度学习方法在公开网站UCI的机器学习知识库的人体行为识别数据集上的识别效果，论证了基于Dropout深度卷积神经网络特征学习方法的有效性。

Comparison of Deep Feature Learning Methods for Human Activity Recognition

In this paper, two kinds of deep feature learning methods are compared for the task of human activity recognition which oriented to smart-phone inertial acceleration sensor data. Compared with the traditional feature engineering method, the deep feature learning method can be trained in end-to-end mode. The network structure is simple and intuitive and it avoids the cumbersome process of feature engineering.The discriminative features are automatically extracted through the deep neural network. In this paper,we compare the performance of two kinds of deep learning methods, including Deep convolution neural network (DCNN) and Long short-term memory network (LSTM) on the UCI human activity recognition data set. From the experiments, DCNN with Dropout demonstrates better performance over the other network models.