A novel attention-based long short-term memory network for human activity recognition based on wearable sensor data
Huihui Yuan · 2024
Wearable sensor-based activity recognition refers to the utilization of sensors worn on the body to collect human motion data. Through deep learning methods, different activities can be classified and recognized. This technology has wide applications in areas like health monitoring, smartwatches, and sports tracking. While deep learning-based Human Activity Recognition (HAR) has made great progress, challenges such as insufficient feature representation and underutilization of spatial and channel information still exist, leading to potential misclassifications. To address these issues, we propose an Attention-based Long Short-Term Memory Convolutional Neural Network framework (ACL-CNN). In this framework, the Convolutional Block Attention Module (CBAM) applies adaptive spatial and channel attention to the extracted features, highlighting important spatial positions and channel correlations to better capture behavioral characteristics. Additionally, the Long Short-Term Memory (LSTM) module processes the time-series data after feature extraction, effectively capturing long-term dependencies in the temporal sequence and enhancing the model's capability to model time-series data. The experiments are carried out on two publicly accessible datasets, namely WISDM and OPPORTUNITY. The results show that our method achieves the best performance on all four evaluation metrics. This validates the effectiveness and superiority of our method in wearable sensor-based activity recognition.