Weijie Sheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Weijie Sheng has a total of 19 co-authors. Here are the top co-authors by number of shared publications:
- Caiyun Ma
State Key Laboratory of Digital Medical Engineering, School of Instrument Science and Engineering, Southeast University, Nanjing, China
3 shared publications - Chengyu Liu
State Key Laboratory of Digital Medical Engineering, School of Instrument Science and Engineering, Southeast University, Nanjing, China
3 shared publications - Jianqing Li
School of Instrument Science and Engineering, Southeast University, Nanjing, China
2 shared publications - Lina Zhao
The State Key Laboratory of Bioelectronics, School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China
2 shared publications Kongbo Zhu
Southeast University, Gulou
1 shared publication- Meicheng Yang
State Key Laboratory of Digital Medical Engineering, School of Instrument Science and Engineering, Southeast University, Nanjing, China
1 shared publication
Latest Publications
- 27 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-11
An Atrial Fibrillation Detection Strategy Based on Self-Supervised Pretraining in Wearable ECG Monitoring
Atrial fibrillation (AF) is an insidious cardiac arrhythmia, with its incidence increasing annually. Timely screening and home-based interventions play a vital role in the effective management of atrial fibrillation (AF). Within the Internet of Medical Things (IoMT) landscape, wearable electrocardiogram (ECG) monitoring devices have been seamlessly integrated to monitor AF. Nonetheless, the substantial influx of ECG signals awaiting annotation and the exorbitant costs associated with employing specialists for manual annotation pose significant hurdles in the development of AF detection systems. Despite the potential utilization of AF detectors trained on open databases, their efficacy in analyzing continuous wearable ECGs remains inadequate. Self-supervised representation learning proficiently characterizes unlabeled data and enhances data utilization without labels. This study proposes an AF detection strategy based on self-supervised pre-training, aiming for optimal AF detection performance with minimal annotation costs. The proposed method employed self-supervised representation learning and pre-training strategy and was validated on four datasets from the 4th China Physiological Signal Challenge (CPSC2021) database, achieving accuracies of 98.13%, 90.11%, 92.63%, and 91.29%. Additionally, validation on 20 wearable ECG recordings yielded mean accuracies of 99.38% and 99.48% for AF and Non-AF on unlabeled data from recordings used for fine-tuning. We achieved mean accuracies of 97.19% and 94.15% for AF and Non-AF on independent recordings. The results demonstrate the proposed method’s reliability for AF detection in wearable ECG monitoring.
- 27 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-11
An Atrial Fibrillation Detection Strategy Based on Self-Supervised Pretraining in Wearable ECG Monitoring
Atrial fibrillation (AF) is an insidious cardiac arrhythmia, with its incidence increasing annually. Timely screening and home-based interventions play a vital role in the effective management of atrial fibrillation (AF). Within the Internet of Medical Things (IoMT) landscape, wearable electrocardiogram (ECG) monitoring devices have been seamlessly integrated to monitor AF. Nonetheless, the substantial influx of ECG signals awaiting annotation and the exorbitant costs associated with employing specialists for manual annotation pose significant hurdles in the development of AF detection systems. Despite the potential utilization of AF detectors trained on open databases, their efficacy in analyzing continuous wearable ECGs remains inadequate. Self-supervised representation learning proficiently characterizes unlabeled data and enhances data utilization without labels. This study proposes an AF detection strategy based on self-supervised pre-training, aiming for optimal AF detection performance with minimal annotation costs. The proposed method employed self-supervised representation learning and pre-training strategy and was validated on four datasets from the 4th China Physiological Signal Challenge (CPSC2021) database, achieving accuracies of 98.13%, 90.11%, 92.63%, and 91.29%. Additionally, validation on 20 wearable ECG recordings yielded mean accuracies of 99.38% and 99.48% for AF and Non-AF on unlabeled data from recordings used for fine-tuning. We achieved mean accuracies of 97.19% and 94.15% for AF and Non-AF on independent recordings. The results demonstrate the proposed method’s reliability for AF detection in wearable ECG monitoring.
- 28 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
An Atrial Fibrillation Detection Strategy in Dynamic ECGs With Significant Individual Differences
Atrial fibrillation (AF) is an insidious disease. Many long-term wearable electrocardiogram (ECG) monitoring devices have been used to monitor AF. The accuracy of detectors used to classify AF/sinus rhythm is already very high on the public database. Due to the significant individual differences and interference from other arrhythmias (e.g., premature beats), the performance of the developed AF detectors can degrade when tested on wearable ECGs. To tackle this, we proposed to use domain-adversarial (DA) learning strategy to minimize feature distribution between the annotated public ECG database (the MIT-BIH AF database) and unlabeled dynamic ECG recordings to improve AF recognition accuracy. DA algorithms based on the shifted window Transformer (DA-ST) and residual neural network (DA-RN) were proposed and validated on the 2021 China Physiological Signal Challenge (CPSC) database including four datasets. The accuracies were 93.85%, 89.78%, 91.93%, and 87.35% on the four datasets when using DA-ST. The corresponding results were 96.67%, 92.25%, 90.58%, and 89.46% when utilizing DA-RN. Importantly, these results demonstrated superior performance compared to the results obtained without DA. The proposed method was validated on 12 wearable long-term recordings, consisting of 4 recordings with premature beats, 4 recordings of AF with premature beats, 2 recordings of sinus rhythms, and 2 recordings of AF. The average results were 98.67% (DA-ST) and 97.89% (DA-RN), proving that the proposed method could provide reliable AF detection for dynamic ECG recordings with significant individual differences.
- 28 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
An Atrial Fibrillation Detection Strategy in Dynamic ECGs With Significant Individual Differences
Atrial fibrillation (AF) is an insidious disease. Many long-term wearable electrocardiogram (ECG) monitoring devices have been used to monitor AF. The accuracy of detectors used to classify AF/sinus rhythm is already very high on the public database. Due to the significant individual differences and interference from other arrhythmias (e.g., premature beats), the performance of the developed AF detectors can degrade when tested on wearable ECGs. To tackle this, we proposed to use domain-adversarial (DA) learning strategy to minimize feature distribution between the annotated public ECG database (the MIT-BIH AF database) and unlabeled dynamic ECG recordings to improve AF recognition accuracy. DA algorithms based on the shifted window Transformer (DA-ST) and residual neural network (DA-RN) were proposed and validated on the 2021 China Physiological Signal Challenge (CPSC) database including four datasets. The accuracies were 93.85%, 89.78%, 91.93%, and 87.35% on the four datasets when using DA-ST. The corresponding results were 96.67%, 92.25%, 90.58%, and 89.46% when utilizing DA-RN. Importantly, these results demonstrated superior performance compared to the results obtained without DA. The proposed method was validated on 12 wearable long-term recordings, consisting of 4 recordings with premature beats, 4 recordings of AF with premature beats, 2 recordings of sinus rhythms, and 2 recordings of AF. The average results were 98.67% (DA-ST) and 97.89% (DA-RN), proving that the proposed method could provide reliable AF detection for dynamic ECG recordings with significant individual differences.
- 6 February 2023
- Published by Cambridge University Press (CUP) in Robotica
- Vol. 41 (5) , 1452-1465
Data augmentation by separating identity and emotion representations for emotional gait recognition
Human-centered intelligent human–robot interaction can transcend the traditional keyboard and mouse and have the capacity to understand human communicative intentions by actively mining implicit human clues (e.g., identity information and emotional information) to meet individuals’ needs. Gait is a unique biometric feature that can provide reliable information to recognize emotions even when viewed from a distance. However, the insufficient amount and diversity of training data annotated with emotions severely hinder the application of gait emotion recognition. In this paper, we propose an adversarial learning framework for emotional gait dataset augmentation, with which a two-stage model can be trained to generate a number of synthetic emotional samples by separating identity and emotion representations from gait trajectories. To our knowledge, this is the first work to realize the mutual transformation between natural gait and emotional gait. Experimental results reveal that the synthetic gait samples generated by the proposed networks are rich in emotional information. As a result, the emotion classifier trained on the augmented dataset is competitive with state-of-the-art gait emotion recognition works.
- 6 February 2023
- Published by Cambridge University Press (CUP) in Robotica
- Vol. 41 (5) , 1452-1465
Data augmentation by separating identity and emotion representations for emotional gait recognition
Human-centered intelligent human–robot interaction can transcend the traditional keyboard and mouse and have the capacity to understand human communicative intentions by actively mining implicit human clues (e.g., identity information and emotional information) to meet individuals’ needs. Gait is a unique biometric feature that can provide reliable information to recognize emotions even when viewed from a distance. However, the insufficient amount and diversity of training data annotated with emotions severely hinder the application of gait emotion recognition. In this paper, we propose an adversarial learning framework for emotional gait dataset augmentation, with which a two-stage model can be trained to generate a number of synthetic emotional samples by separating identity and emotion representations from gait trajectories. To our knowledge, this is the first work to realize the mutual transformation between natural gait and emotional gait. Experimental results reveal that the synthetic gait samples generated by the proposed networks are rich in emotional information. As a result, the emotion classifier trained on the augmented dataset is competitive with state-of-the-art gait emotion recognition works.
- 1 June 2020
- Published by Elsevier in Neurocomputing
- Vol. 395, 86-94
Siamese denoising autoencoders for joints trajectories reconstruction and robust gait recognition
- 1 June 2020
- Published by Elsevier in Neurocomputing
- Vol. 395, 86-94
Siamese denoising autoencoders for joints trajectories reconstruction and robust gait recognition