Wenhui Chen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wenhui Chen has a total of 24 co-authors. Here are the top co-authors by number of shared publications:
- Hoekyung Jung
Computer Engineering Department, Paichai University, Daejeon, Republic of Korea
3 shared publications Huihuang Zhao
College of Computer Science and Technology, Hengyang Normal University, Hengyang, China
3 shared publications- Weijie Wang
College of Computer Science and Technology, Hengyang Normal University, Hengyang 421002, China
3 shared publications - Guanchen Wu
Department of Information Engineering, Guizhou Communications Polytechnic, Guiyang 551400, China
2 shared publications Li Yan
College of Computer Science and Technology, Hengyang Normal University, Hengyang, China
2 shared publications- Mugang Lin
College of Computer Science and Technology, Hengyang Normal University, Hengyang, China
2 shared publications
Latest Publications
- 6 November 2026
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 12 (6) , 065007
NCEL: convolution-enhanced lightweight neural architecture for robust EEG signal decoding in noninvasive brain–computer interfaces
Electroencephalography (EEG) is a cornerstone of brain-computer interfaces and clinical neuro-diagnostics due to its non-invasiveness and high temporal resolution. However, robust EEG decoding remains challenging, particularly for resource-constrained deployments. While lightweight architectures like EEGNet offer favorable parameter efficiency, their reliance on fixed convolution kernels and restricted temporal receptive fields inherently limits their performance in complex, highly variable scenarios. To address these limitations, we propose a Neural Convolution-Enhanced Lightweight network (NCEL) that augments spatiotemporal EEG representations while preserving structural compactness. Specifically, a dynamic neural convolution module (NCM) is designed to learn input-dependent temporal filtering strategies for trial-to-trial variability and session-related distribution shifts.Furthermore, parallel multiscale temporal convolution branches are integrated to concurrently capture short-term local waveform transients and long-interval rhythmic dependencies. Finally, a lightweight channel-attention gating mechanism adaptively recalibrates intermediate representations, amplifying task-relevant neural responses while suppressing noise-dominated channels. Extensive evaluations on five benchmark datasets (BCIC IV-2a, BCIC IV-2b, ERN, SMR, and P300) demonstrate that NCEL achieves a mean classification accuracy of 82.75% on BCIC IV-2b. Notably, NCEL achieves higher mean performance than the EEGNet reference values, with performance gains of 12.53 percentage points on the P300 dataset and 9.54 percentage points on BCIC IV-2b, while maintaining only ∼ 0.008 M parameters. Additional subject-specific session-disjoint evaluations show comparable cross-session performance to EEGNet on BCIC IV-2a and a 1.80-percentage-point mean improvement on BCIC IV-2b. Overall, NCEL provides a parameter-efficient framework for resource-constrained EEG decoding, while broader cross-session generalization requires further validation.
- 6 November 2026
- Published by IOP Publishing in Biomedical Physics & Engineering Express
- Vol. 12 (6) , 065007
NCEL: convolution-enhanced lightweight neural architecture for robust EEG signal decoding in noninvasive brain–computer interfaces
Electroencephalography (EEG) is a cornerstone of brain-computer interfaces and clinical neuro-diagnostics due to its non-invasiveness and high temporal resolution. However, robust EEG decoding remains challenging, particularly for resource-constrained deployments. While lightweight architectures like EEGNet offer favorable parameter efficiency, their reliance on fixed convolution kernels and restricted temporal receptive fields inherently limits their performance in complex, highly variable scenarios. To address these limitations, we propose a Neural Convolution-Enhanced Lightweight network (NCEL) that augments spatiotemporal EEG representations while preserving structural compactness. Specifically, a dynamic neural convolution module (NCM) is designed to learn input-dependent temporal filtering strategies for trial-to-trial variability and session-related distribution shifts.Furthermore, parallel multiscale temporal convolution branches are integrated to concurrently capture short-term local waveform transients and long-interval rhythmic dependencies. Finally, a lightweight channel-attention gating mechanism adaptively recalibrates intermediate representations, amplifying task-relevant neural responses while suppressing noise-dominated channels. Extensive evaluations on five benchmark datasets (BCIC IV-2a, BCIC IV-2b, ERN, SMR, and P300) demonstrate that NCEL achieves a mean classification accuracy of 82.75% on BCIC IV-2b. Notably, NCEL achieves higher mean performance than the EEGNet reference values, with performance gains of 12.53 percentage points on the P300 dataset and 9.54 percentage points on BCIC IV-2b, while maintaining only ∼ 0.008 M parameters. Additional subject-specific session-disjoint evaluations show comparable cross-session performance to EEGNet on BCIC IV-2a and a 1.80-percentage-point mean improvement on BCIC IV-2b. Overall, NCEL provides a parameter-efficient framework for resource-constrained EEG decoding, while broader cross-session generalization requires further validation.
- Open Access
Rabbit: Adaptive Lossless Compression for Floating-Point Time Series via Temporal Locality-Aware Dynamic Encoding
With the advancement of IoT technology, a vast amount of floating-point time series data has emerged, posing significant challenges for data storage and transmission. To address this issue, the efficient compression of floating-point time series data has become increasingly important. In the field of lossless compression where precision loss is not allowed, compression and decompression are a symmetrical and reversible transformation. The optimization of its encoding and decoding strategies remains the current optimal path for lossless compression. Based on the existing lossless compression algorithms for time series, this paper proposes Rabbit, which is a new floating-point time series data stream lossless compression algorithm. This method can perceive the data characteristics and, by leveraging the temporal locality of the time series, predict the branch distribution of the data stream during compression, thereby dynamically encoding the flag bits. This algorithm designs a TOE encoding method specifically for the significant bits to reduce the number of compressed bits. Compared with traditional floating-point compression schemes, its performance has been significantly improved. Experimental evaluations on 28 datasets show that this algorithm consistently outperforms existing methods with an average improvement of 4.15% over the baseline ACTF algorithm. Notably, on datasets such as server31, server34, and server41, the compression ratio can be reduced by up to 43.04%. Additionally, the compression and decompression time metrics have improved by 4.27% and 3.74%, respectively. Overall, Rabbit offers an effective lossless compression approach for floating-point time-series data, improving the compression ratio without compromising encoding/decoding throughput.
Open AccessRabbit: Adaptive Lossless Compression for Floating-Point Time Series via Temporal Locality-Aware Dynamic Encoding
With the advancement of IoT technology, a vast amount of floating-point time series data has emerged, posing significant challenges for data storage and transmission. To address this issue, the efficient compression of floating-point time series data has become increasingly important. In the field of lossless compression where precision loss is not allowed, compression and decompression are a symmetrical and reversible transformation. The optimization of its encoding and decoding strategies remains the current optimal path for lossless compression. Based on the existing lossless compression algorithms for time series, this paper proposes Rabbit, which is a new floating-point time series data stream lossless compression algorithm. This method can perceive the data characteristics and, by leveraging the temporal locality of the time series, predict the branch distribution of the data stream during compression, thereby dynamically encoding the flag bits. This algorithm designs a TOE encoding method specifically for the significant bits to reduce the number of compressed bits. Compared with traditional floating-point compression schemes, its performance has been significantly improved. Experimental evaluations on 28 datasets show that this algorithm consistently outperforms existing methods with an average improvement of 4.15% over the baseline ACTF algorithm. Notably, on datasets such as server31, server34, and server41, the compression ratio can be reduced by up to 43.04%. Additionally, the compression and decompression time metrics have improved by 4.27% and 3.74%, respectively. Overall, Rabbit offers an effective lossless compression approach for floating-point time-series data, improving the compression ratio without compromising encoding/decoding throughput.
- 3 March 2026
- Published by Elsevier in Information Processing & Management
- Vol. 63 (6)
Waved zero streaming: A time series compression framework with adaptive segmentation optimization
- 3 March 2026
- Published by Elsevier in Information Processing & Management
- Vol. 63 (6)
Waved zero streaming: A time series compression framework with adaptive segmentation optimization
- Open Access
- 11 February 2026
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 38 (5)
Dolphin: an adaptive lossless compression algorithm for oscillating floating-point time series
- 11 February 2026
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 38 (5)
Open AccessDolphin: an adaptive lossless compression algorithm for oscillating floating-point time series
- Open Access
UAMS: An Unsupervised Anomaly Detection Method Integrating MSAA and SSPCAB
Anomaly detection methods play a crucial role in automated quality control within modern manufacturing systems. In this context, unsupervised methods are increasingly favored due to their independence from large-scale labeled datasets. However, existing methods present limited multi-scale feature extraction ability and may fail to effectively capture subtle anomalies. To address these challenges, we propose UAMS, a pyramid-structured normalization flow framework that leverages the symmetry in feature recombination to harmonize multi-scale interactions. The proposed framework integrates a Multi-Scale Attention Aggregation (MSAA) module for cross-scale dynamic fusion, as well as a Self-Supervised Predictive Convolutional Attention Block (SSPCAB) for spatial channel attention and masked prediction learning. Experiments on the MVTecAD dataset show that UAMS largely outperforms state-of-the-art unsupervised methods, in terms of detection and localization accuracy, while maintaining high inference efficiency. For example, when comparing UAMS against the baseline model on the carpet category, the AUROC is improved from 90.8% to 94.5%, and AUPRO is improved from 91.0% to 92.9%. These findings validate the potential of the proposed method for use in real industrial inspection scenarios.
Open AccessUAMS: An Unsupervised Anomaly Detection Method Integrating MSAA and SSPCAB
Anomaly detection methods play a crucial role in automated quality control within modern manufacturing systems. In this context, unsupervised methods are increasingly favored due to their independence from large-scale labeled datasets. However, existing methods present limited multi-scale feature extraction ability and may fail to effectively capture subtle anomalies. To address these challenges, we propose UAMS, a pyramid-structured normalization flow framework that leverages the symmetry in feature recombination to harmonize multi-scale interactions. The proposed framework integrates a Multi-Scale Attention Aggregation (MSAA) module for cross-scale dynamic fusion, as well as a Self-Supervised Predictive Convolutional Attention Block (SSPCAB) for spatial channel attention and masked prediction learning. Experiments on the MVTecAD dataset show that UAMS largely outperforms state-of-the-art unsupervised methods, in terms of detection and localization accuracy, while maintaining high inference efficiency. For example, when comparing UAMS against the baseline model on the carpet category, the AUROC is improved from 90.8% to 94.5%, and AUPRO is improved from 91.0% to 92.9%. These findings validate the potential of the proposed method for use in real industrial inspection scenarios.
- Open Access
- 21 April 2025
- Published by MDPI AG in Electronics
- Vol. 14 (8) , 1686
Network Function Placement in Virtualized Radio Access Network with Reinforcement Learning Based on Graph Neural Network
In 5G and beyond 5G networks, function placement is a crucial strategy for enhancing the flexibility and efficiency of the Radio Access Network (RAN). However, demonstrating optimal function splitting and placement to meet diverse user demands remains a significant challenge. The function placement problem is known to be NP-hard, and previous studies have attempted to address it using Deep Reinforcement Learning (DRL) approaches. Nevertheless, many existing methods fail to capture the network state in RANs with specific topologies, leading to suboptimal decision-making and resource allocation. In this paper, we propose a method referred to as GDRL, which is a deep reinforcement learning approach that utilizes graph neural networks to address the functional placement problem. To ensure policy stability, we design a policy gradient algorithm called Graph Proximal Policy Optimization (GPPO), which integrates GNNs into both the actor and critic networks. By incorporating both node and edge features, the GDRL enhances feature extraction from the RAN’s nodes and links, providing richer observational data for decision-making and evaluation. This, in turn, enables more accurate and effective decision outcomes. In addition, we formulate the problem as a mixed-integer nonlinear programming model aimed at minimizing the number of active computational nodes while maximizing the centralization level of the virtualized RAN (vRAN). We evaluate the GDRL across different RAN scenarios with varying node configurations. The results demonstrate that our approach achieves superior network centralization and outperforms several existing methods in overall performance.
- 21 April 2025
- Published by MDPI AG in Electronics
- Vol. 14 (8) , 1686
Open AccessNetwork Function Placement in Virtualized Radio Access Network with Reinforcement Learning Based on Graph Neural Network
In 5G and beyond 5G networks, function placement is a crucial strategy for enhancing the flexibility and efficiency of the Radio Access Network (RAN). However, demonstrating optimal function splitting and placement to meet diverse user demands remains a significant challenge. The function placement problem is known to be NP-hard, and previous studies have attempted to address it using Deep Reinforcement Learning (DRL) approaches. Nevertheless, many existing methods fail to capture the network state in RANs with specific topologies, leading to suboptimal decision-making and resource allocation. In this paper, we propose a method referred to as GDRL, which is a deep reinforcement learning approach that utilizes graph neural networks to address the functional placement problem. To ensure policy stability, we design a policy gradient algorithm called Graph Proximal Policy Optimization (GPPO), which integrates GNNs into both the actor and critic networks. By incorporating both node and edge features, the GDRL enhances feature extraction from the RAN’s nodes and links, providing richer observational data for decision-making and evaluation. This, in turn, enables more accurate and effective decision outcomes. In addition, we formulate the problem as a mixed-integer nonlinear programming model aimed at minimizing the number of active computational nodes while maximizing the centralization level of the virtualized RAN (vRAN). We evaluate the GDRL across different RAN scenarios with varying node configurations. The results demonstrate that our approach achieves superior network centralization and outperforms several existing methods in overall performance.
- 2 February 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 272
Heuristic genetic algorithm parameter optimizer: Making lossless compression algorithms efficient and flexible
- 2 February 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 272
Heuristic genetic algorithm parameter optimizer: Making lossless compression algorithms efficient and flexible
- Open Access
- 1 December 2024
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 36 (10)
ACTF: An efficient lossless compression algorithm for time series floating point data
- 1 December 2024
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 36 (10)
Open AccessACTF: An efficient lossless compression algorithm for time series floating point data
- Open Access
- 29 August 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (17) , 8627
An Optimization Method for Personnel Statistics Based on YOLOv4 + DPAC
Compared to traditional detection methods, image-based flow statistics that determine the number of people in a space are contactless, non-perceptual, and high-speed statistical methods that have broad application prospects and potential economic value in business, education, transportation, and other fields. In this paper, we propose that the distributed probability-adjusted confidence (DPAC) function can optimize the reliability of model prediction according to the actual situation. That is, the reliability can be adjusted using the distribution characteristics of the target in the field of view, and a target can be determined with a confidence level that is greater than 0.5 and more accurately. DPAC can assign different target occurrence probability weights to different regions according to target distribution. Adding the DPAC function to a YOLOv4 network model on the basis of having the target confidence of the YOLOv4 network can reduce or improve confidence according to the target distribution and can then output the final confidence level. Using YOLOv4 + DPAC on the brainwash dataset can improve precision by 0.05% compared to the YOLOv4 model when the target confidence threshold is equal to 0.5; it can improve the recall of the model by 0.12% and the AP of the model by 0.12%. This paper also proposes that the distribution in the DPAC function be obtained based on unsupervised learning and verifies its effectiveness.
- 29 August 2022
- Published by MDPI AG in Applied Sciences
- Vol. 12 (17) , 8627
Open AccessAn Optimization Method for Personnel Statistics Based on YOLOv4 + DPAC
Compared to traditional detection methods, image-based flow statistics that determine the number of people in a space are contactless, non-perceptual, and high-speed statistical methods that have broad application prospects and potential economic value in business, education, transportation, and other fields. In this paper, we propose that the distributed probability-adjusted confidence (DPAC) function can optimize the reliability of model prediction according to the actual situation. That is, the reliability can be adjusted using the distribution characteristics of the target in the field of view, and a target can be determined with a confidence level that is greater than 0.5 and more accurately. DPAC can assign different target occurrence probability weights to different regions according to target distribution. Adding the DPAC function to a YOLOv4 network model on the basis of having the target confidence of the YOLOv4 network can reduce or improve confidence according to the target distribution and can then output the final confidence level. Using YOLOv4 + DPAC on the brainwash dataset can improve precision by 0.05% compared to the YOLOv4 model when the target confidence threshold is equal to 0.5; it can improve the recall of the model by 0.12% and the AP of the model by 0.12%. This paper also proposes that the distribution in the DPAC function be obtained based on unsupervised learning and verifies its effectiveness.