A Wireless Sensor Based Multi-layer Hybrid Deep Learning Model for Highly Correlated Human Activity Recognition

Sensor based human activity recognition has obtained more attentions due to its low-cost, low-data throughput, and immunity to environmental effects. However, traditional work in this field mainly focuses on the recognition of simple and small volume human activities. This work targets complicated, correlated and larger size of human activity recognition. In this paper, a multi-layer hybrid deep learning model is built with convolutional neural networks (CNN) and long short-term memory (LSTM). The multi-layer architecture improves the learning and exploration capacity of local features and temporal dependencies, and the hybrid architecture enriches the diversity for data fusion. In addition, Bayesian optimization is applied to the hybrid model to get the optimal parameters and best performance. The experimental results demonstrate the effectiveness of the proposed model with a recognition rate of 89% for 27 correlated activities. Its performance is better than traditional machine learning and other hybrid deep learning models.