Jianfeng Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jianfeng Wu has a total of 30 co-authors. Here are the top co-authors by number of shared publications:
- Junming Xu
College of Electronic Information, Hangzhou Dianzi University, Hangzhou, China
3 shared publications - Shengying Yang
School of Computer Science and Technology, Zhejiang University of Science and Technology, Hangzhou 310023, China
3 shared publications Banteng Liu
College of Information Science and Technology, Zhejiang Shuren University, Hangzhou, China
2 shared publicationsBingyang Zhu
School of Information Engineering, Huzhou Normal University, Huzhou, China
2 shared publicationsWang Ke
College of Information Science and Technology, Zhejiang Shuren University, Hangzhou, China
2 shared publicationsA. L. Li
Hangzhou Emust Technology Co. Ltd., Hangzhou, China
1 shared publication
Latest Publications
- Open Access
- 29 August 2026
- Published by Springer Nature in Scientific Reports
Automatic sleep stage classification via non-contact BCG and dual-layer modeling
- 29 August 2026
- Published by Springer Nature in Scientific Reports
Open AccessAutomatic sleep stage classification via non-contact BCG and dual-layer modeling
- 28 July 2026
- Published by Springer Nature in Journal of Electronic Materials
- Vol. 55 (10) , 8634-8645
Facile Synthesis of Orientation-Controlled NiCo-LDH Nanosheets on Multilayer Graphene for Supercapacitors
- 28 July 2026
- Published by Springer Nature in Journal of Electronic Materials
- Vol. 55 (10) , 8634-8645
Facile Synthesis of Orientation-Controlled NiCo-LDH Nanosheets on Multilayer Graphene for Supercapacitors
- Open Access
- 29 June 2026
- Published by MDPI AG in Agriculture
- Vol. 16 (13) , 1413
Tomato Leaf Disease Identification via Information-Theoretic Entropy Attention and Hierarchical Feature Alignment
Tomato, as a globally vital economic crop, relies heavily on accurate disease recognition to safeguard food security. However, tomato leaf disease identification constitutes a classic fine-grained visual classification task characterized by minimal inter-class variance, spatially sparse lesion features, and complex background interference. These challenges hinder conventional deep learning models from precisely localizing critical discriminative regions. In response to the aforementioned challenges, we introduce EA-HFA, an innovative framework based on deep neural networks that synergistically integrates an Entropy Attention mechanism alongside a Hierarchical Feature Alignment component. Specifically, the Entropy Attention module leverages information-theoretic entropy to quantify pixel-wise predictive uncertainty, adaptively selecting high-confidence pixels to automatically focus the network on sparse yet highly discriminative lesion features. Concurrently, the Hierarchical Feature Alignment module imposes KL-divergence constraints on the temperature-scaled probability distributions across adjacent network layers, enforcing cross-scale consistency in the localization of discriminative regions. Evaluations conducted on the PlantVillage and AI Challenger 2018 benchmarks reveal that EA-HFA achieves Top-1 accuracies of 99.29% and 97.82%, respectively, yielding performance comparable to established deep learning architectures while maintaining a reasonable computational footprint. Furthermore, qualitative analyses indicate that the model tends to attend to minute lesion-relevant areas, providing a certain level of interpretability for its decision-making process. Thus, EA-HFA holds practical potential as an alternative solution for automated plant disease monitoring in precision farming.
- 29 June 2026
- Published by MDPI AG in Agriculture
- Vol. 16 (13) , 1413
Open AccessTomato Leaf Disease Identification via Information-Theoretic Entropy Attention and Hierarchical Feature Alignment
Tomato, as a globally vital economic crop, relies heavily on accurate disease recognition to safeguard food security. However, tomato leaf disease identification constitutes a classic fine-grained visual classification task characterized by minimal inter-class variance, spatially sparse lesion features, and complex background interference. These challenges hinder conventional deep learning models from precisely localizing critical discriminative regions. In response to the aforementioned challenges, we introduce EA-HFA, an innovative framework based on deep neural networks that synergistically integrates an Entropy Attention mechanism alongside a Hierarchical Feature Alignment component. Specifically, the Entropy Attention module leverages information-theoretic entropy to quantify pixel-wise predictive uncertainty, adaptively selecting high-confidence pixels to automatically focus the network on sparse yet highly discriminative lesion features. Concurrently, the Hierarchical Feature Alignment module imposes KL-divergence constraints on the temperature-scaled probability distributions across adjacent network layers, enforcing cross-scale consistency in the localization of discriminative regions. Evaluations conducted on the PlantVillage and AI Challenger 2018 benchmarks reveal that EA-HFA achieves Top-1 accuracies of 99.29% and 97.82%, respectively, yielding performance comparable to established deep learning architectures while maintaining a reasonable computational footprint. Furthermore, qualitative analyses indicate that the model tends to attend to minute lesion-relevant areas, providing a certain level of interpretability for its decision-making process. Thus, EA-HFA holds practical potential as an alternative solution for automated plant disease monitoring in precision farming.
- Open Access
- 27 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (9) , 1841
TLA-SleepNet: A Transformer–BiLSTM–Attention Network for Automatic Sleep Staging Using Single-Channel Ballistocardiogram Signals
Traditional sleep staging studies typically rely on signals collected using contact-based sensors, which may interfere with the natural sleep state of subjects and thus affect the authenticity and reliability of the recorded data. To address this limitation, this study proposes an automatic sleep staging method based on non-contact single-channel ballistocardiogram (BCG) signals. First, band-pass filtering is applied to the raw BCG signals to separate the heart rate and respiratory components. Heart rate variability (HRV) and respiratory rate variability (RRV) features are then extracted, and mutual information is used to select key feature subsets that exhibit strong correlations with different sleep stages. Considering the complexity and prominent temporal characteristics of real-world sleep data, a temporal modeling network named TLA-SleepNet is constructed to enhance the model’s capability in capturing complex sequential features and improving robustness. Experiments conducted on 10 independent sleep recordings containing a total of 10,614 sleep epochs demonstrate that, under subject non-independent testing conditions with five-fold cross-validation, the proposed method achieves an accuracy of 87.1% in the sleep staging task, with precision, kappa coefficient, and F1-score reaching 92.4%, 81.9%, and 88.7%, respectively. The results indicate that the proposed method can achieve a reliable sleep staging performance without direct contact between sensors and the human body, providing a feasible solution for non-contact sleep monitoring in home-based and mobile healthcare applications.
- 27 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (9) , 1841
Open AccessTLA-SleepNet: A Transformer–BiLSTM–Attention Network for Automatic Sleep Staging Using Single-Channel Ballistocardiogram Signals
Traditional sleep staging studies typically rely on signals collected using contact-based sensors, which may interfere with the natural sleep state of subjects and thus affect the authenticity and reliability of the recorded data. To address this limitation, this study proposes an automatic sleep staging method based on non-contact single-channel ballistocardiogram (BCG) signals. First, band-pass filtering is applied to the raw BCG signals to separate the heart rate and respiratory components. Heart rate variability (HRV) and respiratory rate variability (RRV) features are then extracted, and mutual information is used to select key feature subsets that exhibit strong correlations with different sleep stages. Considering the complexity and prominent temporal characteristics of real-world sleep data, a temporal modeling network named TLA-SleepNet is constructed to enhance the model’s capability in capturing complex sequential features and improving robustness. Experiments conducted on 10 independent sleep recordings containing a total of 10,614 sleep epochs demonstrate that, under subject non-independent testing conditions with five-fold cross-validation, the proposed method achieves an accuracy of 87.1% in the sleep staging task, with precision, kappa coefficient, and F1-score reaching 92.4%, 81.9%, and 88.7%, respectively. The results indicate that the proposed method can achieve a reliable sleep staging performance without direct contact between sensors and the human body, providing a feasible solution for non-contact sleep monitoring in home-based and mobile healthcare applications.
- Open Access
- 18 December 2025
- Published by Springer Nature in Research Square
Automatic Sleep Stage Classification via Non-Contact BCG and Dual-Layer Modeling
Traditional sleep staging studies often rely on contact sensors, which may interfere with natural sleep. Therefore, this study proposes an automatic sleep staging method based on non-contact single channel ballistocardiogram (BCG) signals. Firstly, discrete wavelet transform (DWT) is used to decompose the BCG signal to extract the required heart rate and respiratory components. Subsequently, the mapping relationship between BCG features and sleep staging was established through feature peak detection and further calculation of heart rate variability (HRV) and respiratory variability (RRV). Besides, a Dual-Layer model is introduced to enhance classification accuracy. The five-class classification task is transformed into two three-class problems to improve the distinction between similar sleep stages. Finally, the model was validated on a dataset of 15 independent BCG records (16,660 segments). It achieved an accuracy of 82.9%, improving by 2% to 5% over traditional sleep staging models. It outperformed traditional methods by 2%. This approach overcomes the limitations of sensor contact and offers a more comfortable and accurate method for sleep staging.
- 18 December 2025
- Published by Springer Nature in Research Square
Open AccessAutomatic Sleep Stage Classification via Non-Contact BCG and Dual-Layer Modeling
Traditional sleep staging studies often rely on contact sensors, which may interfere with natural sleep. Therefore, this study proposes an automatic sleep staging method based on non-contact single channel ballistocardiogram (BCG) signals. Firstly, discrete wavelet transform (DWT) is used to decompose the BCG signal to extract the required heart rate and respiratory components. Subsequently, the mapping relationship between BCG features and sleep staging was established through feature peak detection and further calculation of heart rate variability (HRV) and respiratory variability (RRV). Besides, a Dual-Layer model is introduced to enhance classification accuracy. The five-class classification task is transformed into two three-class problems to improve the distinction between similar sleep stages. Finally, the model was validated on a dataset of 15 independent BCG records (16,660 segments). It achieved an accuracy of 82.9%, improving by 2% to 5% over traditional sleep staging models. It outperformed traditional methods by 2%. This approach overcomes the limitations of sensor contact and offers a more comfortable and accurate method for sleep staging.
- Open Access
MetaRes-DMT-AS: A Meta-Learning Approach for Few-Shot Fault Diagnosis in Elevator Systems
Recent advancements in deep learning have spurred significant research interest in fault diagnosis for elevator systems. However, conventional approaches typically require substantial labeled datasets that are often impractical to obtain in real-world industrial environments. This limitation poses a fundamental challenge for developing robust diagnostic models capable of performing reliably under data-scarce conditions. To address this critical gap, we propose MetaRes-DMT-AS (Meta-ResNet with Dynamic Meta-Training and Adaptive Scheduling), a novel meta-learning framework for few-shot fault diagnosis. Our methodology employs Gramian Angular Fields to transform 1D raw sensor data into 2D image representations, followed by episodic task construction through stochastic sampling. During meta-training, the system acquires transferable prior knowledge through optimized parameter initialization, while an adaptive scheduling module dynamically configures support/query sets. Subsequent regularization via prototype networks ensures stable feature extraction. Comprehensive validation using the Case Western Reserve University bearing dataset and proprietary elevator acceleration data demonstrates the framework’s superiority: MetaRes-DMT-AS achieves state-of-the-art few-shot classification performance, surpassing benchmark models by 0.94–1.78% in overall accuracy. For critical few-shot fault categories—particularly emergency stops and severe vibrations—the method delivers significant accuracy improvements of 3–16% and 17–29%, respectively.
Open AccessMetaRes-DMT-AS: A Meta-Learning Approach for Few-Shot Fault Diagnosis in Elevator Systems
Recent advancements in deep learning have spurred significant research interest in fault diagnosis for elevator systems. However, conventional approaches typically require substantial labeled datasets that are often impractical to obtain in real-world industrial environments. This limitation poses a fundamental challenge for developing robust diagnostic models capable of performing reliably under data-scarce conditions. To address this critical gap, we propose MetaRes-DMT-AS (Meta-ResNet with Dynamic Meta-Training and Adaptive Scheduling), a novel meta-learning framework for few-shot fault diagnosis. Our methodology employs Gramian Angular Fields to transform 1D raw sensor data into 2D image representations, followed by episodic task construction through stochastic sampling. During meta-training, the system acquires transferable prior knowledge through optimized parameter initialization, while an adaptive scheduling module dynamically configures support/query sets. Subsequent regularization via prototype networks ensures stable feature extraction. Comprehensive validation using the Case Western Reserve University bearing dataset and proprietary elevator acceleration data demonstrates the framework’s superiority: MetaRes-DMT-AS achieves state-of-the-art few-shot classification performance, surpassing benchmark models by 0.94–1.78% in overall accuracy. For critical few-shot fault categories—particularly emergency stops and severe vibrations—the method delivers significant accuracy improvements of 3–16% and 17–29%, respectively.
- Open Access
- 23 April 2025
- Published by MDPI AG in Nanomaterials
- Vol. 15 (9) , 643
Mechanically Exfoliated Multilayer Graphene-Supported Ni-MOF Parallelogram Nanosheets for Enhanced Supercapacitor Performance
Metal–organic frameworks (MOFs) are regarded as advanced supercapacitor materials owing to their high surface area, redox-active sites, and porosity. However, their insufficient charge carrier mobility remains a critical limitation for practical application. Integrating MOFs with conductive carbon substrates is an effective strategy to break through this limitation. However, conventional carbon materials often require complex preparation methods and pre-activation steps for use in MOF composites. Herein, multilayer graphene (MLG) mechanically exfoliated from expandable graphite is employed as a substrate, and a van der Waals force-assisted chemical deposition method is developed to directly anchor Ni-MOF onto its surface without requiring pre-activation treatment. To optimize the composite, Ni-MOFs with various mass loadings are synthesized on MLG surface. The morphological characteristics and energy storage performance of these composites are thoroughly characterized. Ni-MOF/MLG-0.30 (with a 70.8% Ni-MOF loading on MLG) features a porous stacking structure of well-crystalline Ni-MOF parallelogram nanosheets on MLG, exhibiting optimal electrochemical performance. The composite achieves 1071.4 F·g−1 at 1 A·g−1, and a capacitance retention of 64.9% at the elevated current density of 10 A·g−1. Meanwhile, the composite maintains 63.2% of its initial capacitance after 5000 charge/discharge cycles at 4 A·g−1. A hybrid supercapacitor is fabricated using Ni-MOF/MLG-0.30 cathode and activated carbon anode, delivering 27.9 Wh·kg−1 energy density at 102.5 W·kg−1 power output.
- 23 April 2025
- Published by MDPI AG in Nanomaterials
- Vol. 15 (9) , 643
Open AccessMechanically Exfoliated Multilayer Graphene-Supported Ni-MOF Parallelogram Nanosheets for Enhanced Supercapacitor Performance
Metal–organic frameworks (MOFs) are regarded as advanced supercapacitor materials owing to their high surface area, redox-active sites, and porosity. However, their insufficient charge carrier mobility remains a critical limitation for practical application. Integrating MOFs with conductive carbon substrates is an effective strategy to break through this limitation. However, conventional carbon materials often require complex preparation methods and pre-activation steps for use in MOF composites. Herein, multilayer graphene (MLG) mechanically exfoliated from expandable graphite is employed as a substrate, and a van der Waals force-assisted chemical deposition method is developed to directly anchor Ni-MOF onto its surface without requiring pre-activation treatment. To optimize the composite, Ni-MOFs with various mass loadings are synthesized on MLG surface. The morphological characteristics and energy storage performance of these composites are thoroughly characterized. Ni-MOF/MLG-0.30 (with a 70.8% Ni-MOF loading on MLG) features a porous stacking structure of well-crystalline Ni-MOF parallelogram nanosheets on MLG, exhibiting optimal electrochemical performance. The composite achieves 1071.4 F·g−1 at 1 A·g−1, and a capacitance retention of 64.9% at the elevated current density of 10 A·g−1. Meanwhile, the composite maintains 63.2% of its initial capacitance after 5000 charge/discharge cycles at 4 A·g−1. A hybrid supercapacitor is fabricated using Ni-MOF/MLG-0.30 cathode and activated carbon anode, delivering 27.9 Wh·kg−1 energy density at 102.5 W·kg−1 power output.
- Open Access
- 24 June 2024
- Published by MDPI AG in Electronics
- Vol. 13 (13) , 2472
A Model for Detecting Abnormal Elevator Passenger Behavior Based on Video Classification
In the task of human behavior detection, video classification based on deep learning has become a prevalent technique. The existing models are limited due to an inadequate understanding of behavior characteristics, which restricts their ability to achieve more accurate recognition results. To address this issue, this paper proposes a new model, which is an improvement upon the existing PPTSM model. Specifically, our model employs a multi-scale dilated attention mechanism, which enables the model to integrate multi-scale semantic information and capture characteristic information of abnormal human behavior more effectively. Additionally, to enhance the characteristic information of human behavior, we propose a gradient flow feature information fusion module that integrates high-level semantic features with low-level detail features, enabling the network to extract more comprehensive features. Experiments conducted on an elevator passenger dataset containing four abnormal behaviors (door picking, jumping, kicking, and door blocking) show that the top-1 Acc of our model is improved by 10% compared to the PPTSM model, reaching 95%. Moreover, experiments with four publicly available datasets(UCF24, UCF101, HMDB51, and the Something-Something-v1 dataset) demonstrate that our method achieves results superior to PPTSM by 6.8%, 6.1%, 21.2%, and 3.96%, respectively.
- 24 June 2024
- Published by MDPI AG in Electronics
- Vol. 13 (13) , 2472
Open AccessA Model for Detecting Abnormal Elevator Passenger Behavior Based on Video Classification
In the task of human behavior detection, video classification based on deep learning has become a prevalent technique. The existing models are limited due to an inadequate understanding of behavior characteristics, which restricts their ability to achieve more accurate recognition results. To address this issue, this paper proposes a new model, which is an improvement upon the existing PPTSM model. Specifically, our model employs a multi-scale dilated attention mechanism, which enables the model to integrate multi-scale semantic information and capture characteristic information of abnormal human behavior more effectively. Additionally, to enhance the characteristic information of human behavior, we propose a gradient flow feature information fusion module that integrates high-level semantic features with low-level detail features, enabling the network to extract more comprehensive features. Experiments conducted on an elevator passenger dataset containing four abnormal behaviors (door picking, jumping, kicking, and door blocking) show that the top-1 Acc of our model is improved by 10% compared to the PPTSM model, reaching 95%. Moreover, experiments with four publicly available datasets(UCF24, UCF101, HMDB51, and the Something-Something-v1 dataset) demonstrate that our method achieves results superior to PPTSM by 6.8%, 6.1%, 21.2%, and 3.96%, respectively.
- Open Access
- 21 October 2022
- Published by MDPI AG in Nanomaterials
- Vol. 12 (20) , 3697
Nanowires-Assembled TiO2 Nanorods Anchored on Multilayer Graphene for High-Performance Anodes of Lithium-Ion Batteries
Multilayer graphene (MLG) prepared via ultrasonic exfoliation has many advantages such as its low-cost and defect-free nature, high electronic conductivity, and large specific surface area, which make it an apt conductive substrate for TiO2 composites. To synthesize graphene/TiO2 hybrids, traditional methods that greatly depend on the chemical bond of oxygen-containing functional groups on graphene with titanium cations are not applicable due to the absence of these functional groups on MLG. In this work, a facile chemical method is developed to directly deposit TiO2 on the MLG surface without the introduction of chemically active groups. With this method, four types of TiO2 materials, that is pure anatase TiO2 nanoparticles, a mixture of anatase TiO2 nanoparticles and rutile TiO2 nanoflowers, pure rutile TiO2 nanoflowers, and pure rutile TiO2 nanorods, are homogeneously anchored on the MLG surface by controlling the amount of HCl in the reactant. Interestingly, the rutile TiO2 nanorods in the TiO2/MLG composite are assembled by many TiO2 nanowires with an ultra-small diameter and ultra-long length, which provides a better synergetic effect for high performances as LIB anodes than other composites. A specific capacity of 631.4 mAh g−1 after 100 cycles at a current density of 100 mA g−1 is delivered, indicating it to be a valuable LIB anode material with low cost and high electrochemical performances.
- 21 October 2022
- Published by MDPI AG in Nanomaterials
- Vol. 12 (20) , 3697
Open AccessNanowires-Assembled TiO2 Nanorods Anchored on Multilayer Graphene for High-Performance Anodes of Lithium-Ion Batteries
Multilayer graphene (MLG) prepared via ultrasonic exfoliation has many advantages such as its low-cost and defect-free nature, high electronic conductivity, and large specific surface area, which make it an apt conductive substrate for TiO2 composites. To synthesize graphene/TiO2 hybrids, traditional methods that greatly depend on the chemical bond of oxygen-containing functional groups on graphene with titanium cations are not applicable due to the absence of these functional groups on MLG. In this work, a facile chemical method is developed to directly deposit TiO2 on the MLG surface without the introduction of chemically active groups. With this method, four types of TiO2 materials, that is pure anatase TiO2 nanoparticles, a mixture of anatase TiO2 nanoparticles and rutile TiO2 nanoflowers, pure rutile TiO2 nanoflowers, and pure rutile TiO2 nanorods, are homogeneously anchored on the MLG surface by controlling the amount of HCl in the reactant. Interestingly, the rutile TiO2 nanorods in the TiO2/MLG composite are assembled by many TiO2 nanowires with an ultra-small diameter and ultra-long length, which provides a better synergetic effect for high performances as LIB anodes than other composites. A specific capacity of 631.4 mAh g−1 after 100 cycles at a current density of 100 mA g−1 is delivered, indicating it to be a valuable LIB anode material with low cost and high electrochemical performances.