Fei Deng
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
Fei Deng has a total of 43 co-authors. Here are the top co-authors by number of shared publications:
- Golam Saklain
College of Information Engineering, Kunming University, Kunming, Yunnan, China
4 shared publications Kaiguo Qian
Kunming University
4 shared publicationsShibin Liang
Research and Development Center, Power Research Institute of Yunnan Electric Power Testing and Research Institute (Group) Co., Ltd., Kunming, Yunnan, China
4 shared publicationsShikai Shen
College of Information Engineering, Kunming University, Kunming, Yunnan, China
3 shared publications- Xisong Miao
College of Information Engineering, Kunming University, Kunming, Yunnan, China
3 shared publications - Bowen Zheng
School of Information Engineering, Kunming University, Kunming, Yunnan, China
2 shared publications
Latest Publications
- Open Access
- 3 July 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3956
Stability analysis for a recurrent sigma‒pi‒sigma neural network based on a batch gradient algorithm with L2 regularization
Higher-order neural networks (HONNs) have more powerful nonlinear mapping capabilities than traditional feed-forward neural networks do. The recurrent sigma-pi-sigma neural network (RSPSNN), a multilayer higher-order neural network, suffers from slow convergence and poor generalizability when trained using a traditional gradient learning algorithm. To overcome these drawbacks, this work presents a method of accelerating the training of the RSPSNN on the basis of L2 regularization, and the results show that the L2 regularity term can effectively accelerate the convergence of the network during the training process. In addition, we rigorously demonstrate the convergence stability of the proposed algorithm, and the performance of the algorithm is evaluated via classification and approximation experiments, which confirm the theoretical results of this article.
- 3 July 2026
- Published by PeerJ in PeerJ Computer Science
- Vol. 12, e3956
Open AccessStability analysis for a recurrent sigma‒pi‒sigma neural network based on a batch gradient algorithm with L2 regularization
Higher-order neural networks (HONNs) have more powerful nonlinear mapping capabilities than traditional feed-forward neural networks do. The recurrent sigma-pi-sigma neural network (RSPSNN), a multilayer higher-order neural network, suffers from slow convergence and poor generalizability when trained using a traditional gradient learning algorithm. To overcome these drawbacks, this work presents a method of accelerating the training of the RSPSNN on the basis of L2 regularization, and the results show that the L2 regularity term can effectively accelerate the convergence of the network during the training process. In addition, we rigorously demonstrate the convergence stability of the proposed algorithm, and the performance of the algorithm is evaluated via classification and approximation experiments, which confirm the theoretical results of this article.
- 30 April 2026
- Published by Oxford University Press (OUP) in The Computer Journal
- Vol. 69 (10) , 1602-1617
PSAF-Net: a position-aware sparse encoding and structure-aware fusion network for open-set network intrusion detection
In cybersecurity, intrusion detection systems face growing challenges due to increasingly heterogeneous attack patterns and evolving behaviors, particularly in the recognition of unknown threats. Existing approaches show limited generalization and unstable decision boundaries due to insufficient modeling of structural divergence, distributional shift, inadequate responsiveness to abrupt local anomalies, and weak multi-scale feature fusion capabilities. To address these issues, we propose a position-aware sparse encoding and structure-aware fusion network (PSAF-Net). Specifically, PSAF-Net incorporates a learnable position-aware encoding module to embed temporal positional cues, improving the modeling of heterogeneous communication structures and behavioral trends. It then introduces a sparsity-enhanced encoding module guided by local variation rates, using sliding-window-based attention to highlight abrupt changes and suppress redundant activations, thereby improving sensitivity to fine-grained anomalies. Finally, a temporal structure-aware feature modeling module is designed to dynamically integrate multi-receptive-field features through a scale-aware guided attention feature fusion mechanism and a cross-scale structure-aware gated fusion, strengthening the model’s capacity to discriminate under diverse attack modes and shifting distributions. Extensive experiments on CIC-IDS2017, UNSW-NB15, and ToN-IoT datasets demonstrate that PSAF-Net significantly outperforms state-of-the-art methods in unknown threat detection, exhibiting superior detection accuracy and open-set generalization.
- 30 April 2026
- Published by Oxford University Press (OUP) in The Computer Journal
- Vol. 69 (10) , 1602-1617
PSAF-Net: a position-aware sparse encoding and structure-aware fusion network for open-set network intrusion detection
In cybersecurity, intrusion detection systems face growing challenges due to increasingly heterogeneous attack patterns and evolving behaviors, particularly in the recognition of unknown threats. Existing approaches show limited generalization and unstable decision boundaries due to insufficient modeling of structural divergence, distributional shift, inadequate responsiveness to abrupt local anomalies, and weak multi-scale feature fusion capabilities. To address these issues, we propose a position-aware sparse encoding and structure-aware fusion network (PSAF-Net). Specifically, PSAF-Net incorporates a learnable position-aware encoding module to embed temporal positional cues, improving the modeling of heterogeneous communication structures and behavioral trends. It then introduces a sparsity-enhanced encoding module guided by local variation rates, using sliding-window-based attention to highlight abrupt changes and suppress redundant activations, thereby improving sensitivity to fine-grained anomalies. Finally, a temporal structure-aware feature modeling module is designed to dynamically integrate multi-receptive-field features through a scale-aware guided attention feature fusion mechanism and a cross-scale structure-aware gated fusion, strengthening the model’s capacity to discriminate under diverse attack modes and shifting distributions. Extensive experiments on CIC-IDS2017, UNSW-NB15, and ToN-IoT datasets demonstrate that PSAF-Net significantly outperforms state-of-the-art methods in unknown threat detection, exhibiting superior detection accuracy and open-set generalization.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 10686-10701
UAV Trajectory, Power, and User-Association Optimization for Covert Communications in ICRNs
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Cognitive Communications and Networking
- Vol. 12, 10686-10701
UAV Trajectory, Power, and User-Association Optimization for Covert Communications in ICRNs
- Open Access
- 19 December 2025
- Published by Kaunas University of Technology (KTU) in Information Technology And Control
- Vol. 54 (4) , 1248-1258
A Novel Method of Object Detection Based on YOLOV8 and PixelAttention in Transmission Lines
In recent years, a large amount of debris has appeared on power transmission lines, affecting circuit power supply and endangering human safety. At present, deep learning based object detection algorithms have made continuous progress, but they still cannot meet the practical application requirements of real-time performance. It is necessary to further reduce model complexity and improve detection speed. Therefore, a transmission line debris recognition based on fusion point attention improved YOLOV8 is proposed. The model uses the original YOLOV8 model as the base model and adds PixelAttention in detection to enhance the model's receptive field and multi-scale object perception, thereby improving the model detection accuracy. The experimental results show that the average accuracy of the original model is only 92.3%, while the average accuracy of the improved model has increased to 94%.
- 19 December 2025
- Published by Kaunas University of Technology (KTU) in Information Technology And Control
- Vol. 54 (4) , 1248-1258
Open AccessA Novel Method of Object Detection Based on YOLOV8 and PixelAttention in Transmission Lines
In recent years, a large amount of debris has appeared on power transmission lines, affecting circuit power supply and endangering human safety. At present, deep learning based object detection algorithms have made continuous progress, but they still cannot meet the practical application requirements of real-time performance. It is necessary to further reduce model complexity and improve detection speed. Therefore, a transmission line debris recognition based on fusion point attention improved YOLOV8 is proposed. The model uses the original YOLOV8 model as the base model and adds PixelAttention in detection to enhance the model's receptive field and multi-scale object perception, thereby improving the model detection accuracy. The experimental results show that the average accuracy of the original model is only 92.3%, while the average accuracy of the improved model has increased to 94%.
- 13 November 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal of Pattern Recognition and Artificial Intelligence
- Vol. 40 (01)
Stability, Monotonicity, and Necessity Analysis of Gradient Method with Penalty Terms for Dynamic Ridge Polynomial Neural Network
In recent years, high-order neural networks (HONNs) have gained widespread usage due to their excellent approximation ability, incorporating both ∑ neurons and ∏ neurons. The dynamic ridge polynomial neural network (DRPNN), as a type of HONN, has been highly valued. However, during the training process of the DRPNN, slow convergence occurs due to the presence of unknown factors. Given this, we propose a method to enhance the DRPNN convergence speed by incorporating a penalty term into the error function and training it using the gradient method. The proposed method significantly improves the convergence speed during network training. To explain the training characteristics of the proposed method, three theorems, namely stability, monotonicity, and necessity, are theoretically proven. More precisely, the whole network is stable over time based on the ultimate training results. The loss function monotonically decreases. Regarding the necessity, the ultimate training results must be an optimal solution. Furthermore, under the condition of the same iteration steps and error accuracy, experiments concerning function simulation and image simulation are advanced to verify the theoretical effectiveness of the proposed method, with the advanced analysis of the corresponding generalizability being discussed.
- 13 November 2025
- Published by World Scientific Pub Co Pte Ltd in International Journal of Pattern Recognition and Artificial Intelligence
- Vol. 40 (01)
Stability, Monotonicity, and Necessity Analysis of Gradient Method with Penalty Terms for Dynamic Ridge Polynomial Neural Network
In recent years, high-order neural networks (HONNs) have gained widespread usage due to their excellent approximation ability, incorporating both ∑ neurons and ∏ neurons. The dynamic ridge polynomial neural network (DRPNN), as a type of HONN, has been highly valued. However, during the training process of the DRPNN, slow convergence occurs due to the presence of unknown factors. Given this, we propose a method to enhance the DRPNN convergence speed by incorporating a penalty term into the error function and training it using the gradient method. The proposed method significantly improves the convergence speed during network training. To explain the training characteristics of the proposed method, three theorems, namely stability, monotonicity, and necessity, are theoretically proven. More precisely, the whole network is stable over time based on the ultimate training results. The loss function monotonically decreases. Regarding the necessity, the ultimate training results must be an optimal solution. Furthermore, under the condition of the same iteration steps and error accuracy, experiments concerning function simulation and image simulation are advanced to verify the theoretical effectiveness of the proposed method, with the advanced analysis of the corresponding generalizability being discussed.
- Open Access
- 5 October 2025
- Published by Springer Nature in Journal of Engineering and Applied Science
- Vol. 72 (1) , 1-16
Decision tree-based fault diagnosis system for distribution network fault indicators
As a widely used fault detection equipment in distribution networks, the distribution network fault indicator plays an important role in timely fault judgment and power supply recovery. Therefore, when designing distribution network lines, the quality of distribution network fault indicators is an important factor affecting the fault judgment of the distribution network. This paper discloses a fault diagnosis method and system for the distribution network fault indicator based on the decision tree learning algorithm, which enhances the real historical data through generative adversarial network (GAN), generates the historical fault data of the distribution network fault indicator and the working operation status of each module of the fault indicator, and applies the decision tree algorithm to classify faults based on the enhanced data, achieving a diagnostic accuracy of 98%. The results show that the application of this method to the fault diagnosis of the distribution network can carry out automatic fault diagnosis and analysis, which greatly saves the labor cost of fault indicator maintenance.
- 5 October 2025
- Published by Springer Nature in Journal of Engineering and Applied Science
- Vol. 72 (1) , 1-16
Open AccessDecision tree-based fault diagnosis system for distribution network fault indicators
As a widely used fault detection equipment in distribution networks, the distribution network fault indicator plays an important role in timely fault judgment and power supply recovery. Therefore, when designing distribution network lines, the quality of distribution network fault indicators is an important factor affecting the fault judgment of the distribution network. This paper discloses a fault diagnosis method and system for the distribution network fault indicator based on the decision tree learning algorithm, which enhances the real historical data through generative adversarial network (GAN), generates the historical fault data of the distribution network fault indicator and the working operation status of each module of the fault indicator, and applies the decision tree algorithm to classify faults based on the enhanced data, achieving a diagnostic accuracy of 98%. The results show that the application of this method to the fault diagnosis of the distribution network can carry out automatic fault diagnosis and analysis, which greatly saves the labor cost of fault indicator maintenance.
- Open Access
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 144-151
Binary Image Classification Using KNN Optimized with Mahalanobis Distance
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 144-151
Open AccessBinary Image Classification Using KNN Optimized with Mahalanobis Distance
- Open Access
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 324-330
A redundancy data discrimination method based on inverse matrix about student mark
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 324-330
Open AccessA redundancy data discrimination method based on inverse matrix about student mark
- Open Access
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 126-131
Exploration of News Categorization Techniques Based on CART Decision Trees
- 18 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 126-131
Open AccessExploration of News Categorization Techniques Based on CART Decision Trees
- Open Access
- 6 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 637-643
Sentiment analysis of microblog data based on TextCNN
- 6 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 637-643
Open AccessSentiment analysis of microblog data based on TextCNN