Abstract:To improve the accuracy of human activity recognition (HAR) based on body area network (BAN), a novel spatio-temporal network combining multi-channel convolutional neural network (CNN) with graph convolutional neural network (GCN) is proposed in this paper. Based on BAN including multi-sensors, our model takes advantage of graph topological structure to represent human activity pattern. The multi-channel CNN is firstly constructed to greatly enhance learning ability to capture local representative feature. This benefits to exploit the most representative temporal activity features from time-series sensor data, overcoming the limitation of a single CNN with poor learning ability. And then the GCN with excellect extraction ability to local spatial feature is employed to discover the most representative spatial activity feature associated with the extracted temporal activity features in non-Euclidean space. This could gather the most valuable spatio-temporal dynamic activity feature hidden in human activity for HAR with high-generalization. The open wearable action recognition database dataset including all 13 actiivty patterns is used to evaluate the effectiveness of our proposed model. The experimental results showed that, with the limited sample data, our model can achieve the best generalization (96.01% accuracy, 95.39% precision, and 96.51% recall) to accurately detect all 13 activity patterns when compared with the recent relevant models. This suggested that our technique could gain surperior spatio-temporal activity representation learning capability in Euclidean and non-Euclidean space based on the limited sample data than the compared models. Our study may provide an effective technical solution for HAR wuth high quality.

Improving Human Activity Recognition Via Graph Attention Network with Linear Discriminant Analysis and Residual Learning

Beyond the Gates of Euclidean Space: Temporal-Discrimination-Fusions and Attention-based Graph Neural Network for Human Activity Recognition

Attention-based LSTM Network for Wearable Human Activity Recognition

Deep Transfer Learning with Graph Neural Network for Sensor-Based Human Activity Recognition

GANCDE: Neural networks based on graphs and attention neural control differential equations for human activity recognition

MAG-Res2Net: a novel deep learning network for human activity recognition

Leveraging spatio-temporal features using graph neural networks for human activity recognition

Multi-STMT: Multi-Level Network for Human Activity Recognition Based on Wearable Sensors

Skeleton-Based Action Recognition with Spatial-Structural Graph Convolution

Human Activity Recognition Based on Residual Network and BiLSTM

GRU with Dual Attentions for Sensor-Based Human Activity Recognition

Multihead-Res-SE Residual Network with Attention for Human Activity Recognition

HAR-GCNN: Deep Graph CNNs for Human Activity Recognition From Highly Unlabeled Mobile Sensor Data

Multi-DGI: Multi-head Pooling Deep Graph Infomax for Human Activity Recognition

Augmented Adversarial Learning for Human Activity Recognition with Partial Sensor Sets

A Novel Spatio-Temporal Network of Multi-channel CNN and GCN for Human Activity Recognition Based on BAN

Deep Residual Bidir-LSTM for Human Activity Recognition Using Wearable Sensors

Improving Human Activity Recognition With Wearable Sensors Through BEE: Leveraging Early Exit and Gradient Boosting

Domain Adaptation for Sensor-Based Human Activity Recognition with a Graph Convolutional Network

Residual deep gated recurrent unit-based attention framework for human activity recognition by exploiting dilated features

A Semisupervised Recurrent Convolutional Attention Model for Human Activity Recognition