30 July 2024 in IEEE Transactions on Industrial Informatics
Inertial sensors, as motion-sensing components, are widely used in inertial navigation, aerospace, and consumer electronics. Their wide applications and severe errors form a sharp contradiction, which attracts considerable attention. Wavelet is hailed as the mathematical signal microscope due to the diverse wavelet basis functions. However, complicated noise types and application scenarios of inertial sensors make selecting wavelet basis perplexing. To this end, we propose a wavelet encoding network (WENet), which intelligently selects the appropriate wavelet for variable inertial signals by representing wavelet characteristics through the devised category representation mechanism (CRM). Furthermore, CRM introduces a feature supervision effect, which imposes interpretability on a black-box network and forces it to provide direct supervision for other network structures. This supervision strategy is closer and more effective than the supervision provided by the output end, which is far and needs to go through backpropagation. The proposed WENet has the reliability of the model-driven method and the flexibility of the data-driven method. As a weakly supervised method, the WENet achieves the best performance among all signal improvement methods, including fully supervised ones. After being enhanced by WENet, the low-cost sensor signal can perform accurate spatial trajectory reconstruction, which was once considered an impossible task.
10 July 2025 in IEEE Transactions on Industrial Informatics
In an increasingly digital world, there is an unprecedented demand in intelligent society for biometric recognition systems that balance security with convenience. Traditional methods, such as passwords and PINs, are vulnerable to breaches and impose the burden of remembering multiple credentials. To this end, we propose a handwriting verification technology leveraging inertial sensors embedded in wearable devices (e.g., smartphones). Handwriting biometrics offer enhanced security as unique writing patterns are inherently resistant to replication and forgery. However, this approach faces two critical challenges: Identity verification independent of written content, and generalizability to unseen writers during deployment. We therefore devise a feature-guided zero-shot learning (FGZSL) framework, which constructs a unique feature vector for each sample and then performs identity verification by comparing these feature vectors. For the first challenge, we design an aim focuser, a module that filters out irrelevant content from the data features, allowing the FGZSL to focus on identity-specific information rather than writing content. For the second challenge, we design a plug-and-play Rényi-entropy-based representation regularization, which constructs informative and discriminative features for seen categories during training. These features serve as bases for representing unseen categories during testing. We contribute the first inertial identity detection dataset, publicly available on GitHub, containing 39 800 training samples and 10 000 test samples. Extensive experiments demonstrate that the FGZSL framework outperforms existing methods in both seen and unseen categories, but also sets a new standard for secure and reliable identity authentication using inertial sensors.