A Novel Separable Convolution Neural Network For Human Activity Recognition

Résumé: The issue with the time series classification arises in several human applications such as healthcare, industrial monitoring and cybersecurity. Recently, various methods have been developed in order to deal with this matter. In this paper, a novel deep learning-based model for human activity recognition is developed. The proposal examines deeply the training phase in which the acceleration metric is considered by exploring all components of the model. To this end, the architecture of the Convolutional Neural Network (CNN) is studied: a) first, we employ a separable CNN, where we integrate a particular filter model for the depthwise convolution; b) second, we combine the extracted features with the handcrafted features. The proposed classifier is evaluated using a human activity recognition dataset and compared to a set of recent works. The obtained results show that our model outperforms the compared methods under various metrics.

Publié dans la revue: