基于卷积神经网络-双向长短期记忆网络的人体活动识别方法

针对人体活动传感器数据的时序性特点，以及当前机器学习算法过度依赖手工特征提取的问题，提出了一种融合卷积神经网络和双向长短期记忆网络的深度学习模型（CNN-BiLSTM）进行人体活动识别（HAR）。首先对人体活动数据进行样本分割，然后采用卷积神经网络（CNN）自动提取人体活动数据的特征，再通过双向长短期记忆网络（BiLSTM）学习人体活动数据特征在时间序列上前后两个方向的相关性，最后利用softmax分类器实现对人体活动分类。DaLiAc公开数据集上的仿真实验结果表明，基于CNN-BiLSTM网络的人体活动识别方法对13种人体活动的识别准确率达到了97.7%，与仅具备时间特征学习的LSTM网络和BiLSTM网络相比，具有更好的识别分类效果。

Human Activity Recognition Method Based on CNN-BiLSTM

Aiming at the time series characteristics of human activity sensor data and the problem that current machine learning algorithms rely excessively on manual feature extraction, a deep learning model (CNN-BiLSTM) fused with convolutional neural network and bidirectional long short term memory network is proposed for Human Activity Recognition (HAR). Firstly, the human activity data is segmented, and then the convolutional neural network (CNN) is used to automatically extract the features of human activity data. Secondly, the bidirectional long short term memory network (BiLSTM) is used to learn the correlation of human activity data features in the two directions of time series. Finally, the softmax classifier is used to realize the classification of human activities. The simulation experiment results on the DaLiAc public dataset show that the human activity recognition method based on the CNN-BiLSTM network has an accuracy of 97.7% for the recognition of 13 human activities. Compared with the LSTM network and the BiLSTM network that only have time feature learning, CNN-BiLSTM network has a better recognition and classification effect.