Enhancing Kinetic Energy Harvesting-based Human Activity Recognition with Deep Learning and Data Augmentation

Kinetic energy harvesting technologies have been progressively used to power wearable devices and to sense the context through energy generation patterns. However, detecting human activities with signals from kinetic harvesters still needs improvement due to the use of approaches based on handcrafted features and the overfitting to device location or subjects. Hence, in this paper we present a deep learning architecture that leverages the feature extraction capability of the convolutional neural networks and the construction of the temporal sequences of recurrent neural networks to improve existing classification results. This architecture outperforms existing approaches of kinetic harvesting-based human activity recognition by 17% of accuracy. Additionally, we demonstrate how our architecture maintains margins up to 15% compared to other classifiers when the signals are affected by downsampling or length changes. To provide sufficient data for the deep learning classifier, we propose three data augmentation methods to increase intraclass variance simulating new users performing the same activities. Finally, given the dependency of kinetic harvesting signals on device location and subjects, we employ transfer learning to develop location-independent and subject-independent models. The combination of transfer learning and data augmentation helps to increase classification performance by 27% when the device location is changed and 44% when the data comes from a new subject.