Yuefeng Cen
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
Yuefeng Cen has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
- Gang Cen
School of Information and Electronic Engineering, Zhejiang University of Science and Technology, Hangzhou, China
13 shared publications - Cheng Zhao
School of Economics, Zhejiang University of Technology, Hangzhou, China
4 shared publications - Gangyi Zhang
Zhejiang University of Science and Technology
3 shared publications - Zeping Yang
Zhejiang University of Science and Technology
3 shared publications - Zhiqi Jin
Zhejiang University of Science and Technology
3 shared publications Cen Gang
School of Information and Electronic Engineering, Zhejiang University of Science & Technology, Hangzhou, China
2 shared publications
Latest Publications
- Open Access
- 17 September 2026
- Published by Elsevier in SSRN Electronic Journal
A UAV-borne edge computing system for early tea anthracnose detection based on a frequency-aware lightweight model
Tea anthracnose, a prevalent disease of Longjing 43—a cultivar extensively grown in eastern China—severely compromises leaf quality and yield. Its frequent infection of tender leaves during the harvest season makes early and accurate diagnosis critical for timely disease management and reduced economic losses. Yet early-stage infected leaf textures nearly indistinguishable from healthy backgrounds, so conventional visual detection often fails. In low-altitude UAV inspections, heavy models cannot meet the real-time constraints of edge computing, whereas conventional lightweight models rely on aggressive downsampling that over-smooths the high-frequency textures of tiny lesions, thereby limiting early recognition.To address these limitations, we propose a UAV-borne edge computing system driven by a frequency-aware lightweight model. The framework adopts an inverted "Pruning-then-Tuning" architecture. First, spatial and channel reconstruction techniques suppress background redundancy to reduce computational cost. Second, frequency-preservingg Haar wavelet downsampling replaces traditional strided convolutions to preserve high-frequency details. The core innovation—the Spatial-Wavelet Learnable Fusion Tuner (SwlFT)—integrates learnable wavelet transforms with an adaptive spatial-frequency gating mechanism, effectively enhancing weak pathological textures obscured by background noise.To bridge algorithmic design with field application, we constructed a UAV-borne optical pod utilizing the Jetson Orin Nano. Experiments demonstrate that the proposed model improves mean Average Precision (mAP@50) by 5.5% over the baseline while reducing FLOPs by 14.6%. Validated across multi-site field inspections comprising 131 minutes of flight data, the hardware-software co-designed system sustained a stable 32.40 FPS without thermal throttling. This framework provides an accurate and field-deployable solution for precision agriculture.
- 17 September 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessA UAV-borne edge computing system for early tea anthracnose detection based on a frequency-aware lightweight model
Tea anthracnose, a prevalent disease of Longjing 43—a cultivar extensively grown in eastern China—severely compromises leaf quality and yield. Its frequent infection of tender leaves during the harvest season makes early and accurate diagnosis critical for timely disease management and reduced economic losses. Yet early-stage infected leaf textures nearly indistinguishable from healthy backgrounds, so conventional visual detection often fails. In low-altitude UAV inspections, heavy models cannot meet the real-time constraints of edge computing, whereas conventional lightweight models rely on aggressive downsampling that over-smooths the high-frequency textures of tiny lesions, thereby limiting early recognition.To address these limitations, we propose a UAV-borne edge computing system driven by a frequency-aware lightweight model. The framework adopts an inverted "Pruning-then-Tuning" architecture. First, spatial and channel reconstruction techniques suppress background redundancy to reduce computational cost. Second, frequency-preservingg Haar wavelet downsampling replaces traditional strided convolutions to preserve high-frequency details. The core innovation—the Spatial-Wavelet Learnable Fusion Tuner (SwlFT)—integrates learnable wavelet transforms with an adaptive spatial-frequency gating mechanism, effectively enhancing weak pathological textures obscured by background noise.To bridge algorithmic design with field application, we constructed a UAV-borne optical pod utilizing the Jetson Orin Nano. Experiments demonstrate that the proposed model improves mean Average Precision (mAP@50) by 5.5% over the baseline while reducing FLOPs by 14.6%. Validated across multi-site field inspections comprising 131 minutes of flight data, the hardware-software co-designed system sustained a stable 32.40 FPS without thermal throttling. This framework provides an accurate and field-deployable solution for precision agriculture.
- 1 May 2026
- Published by Springer Nature
- p. 181-193
The Design and Development of the AI Interviewer System
- 1 May 2026
- Published by Springer Nature
- p. 181-193
The Design and Development of the AI Interviewer System
- 1 May 2026
- Published by Springer Nature
- p. 140-153
Design and Development of AIGC-Based Intelligent Teaching Assistance System
- 1 May 2026
- Published by Springer Nature
- p. 140-153
Design and Development of AIGC-Based Intelligent Teaching Assistance System
- 3 December 2025
- Published by Springer Nature in International Journal of Data Science and Analytics
- Vol. 21 (1) , 48
MAWT: a market-contextual adaptive wavelet transformer for stock forecasting
- 3 December 2025
- Published by Springer Nature in International Journal of Data Science and Analytics
- Vol. 21 (1) , 48
MAWT: a market-contextual adaptive wavelet transformer for stock forecasting
- Open Access
- 31 May 2025
- Published by Springer Nature in Discover Applied Sciences
- Vol. 7 (6) , 1-16
EGRec: a MOOCs course recommendation model based on knowledge graphs
- 31 May 2025
- Published by Springer Nature in Discover Applied Sciences
- Vol. 7 (6) , 1-16
Open AccessEGRec: a MOOCs course recommendation model based on knowledge graphs
- Open Access
- 14 April 2025
- Published by Springer Nature in Research Square
EGRec: A MOOCs Course Recommendation Model Based on Knowledge Graphs
Massive Open Online Courses (MOOCs) provide abundant learning resources but also overwhelm learners with their sheer volume, leading to challenges such as data sparsity and cold-start issues in conventional recommendation systems. To address these challenges, we propose EGRec, a novel course recommendation model that combines knowledge graphs and Heterogeneous Graph Attention Networks to improve recommendation precision, diversity, and relevance. By integrating multimodal data, EGRec captures intricate semantic relationships between courses and knowledge points, enabling personalized and context-sensitive recommendations. Extensive experiments on real MOOCs datasets demonstrate that EGRec significantly outperforms traditional models, highlighting its potential to enhance tailored learning experiences.
- 14 April 2025
- Published by Springer Nature in Research Square
Open AccessEGRec: A MOOCs Course Recommendation Model Based on Knowledge Graphs
Massive Open Online Courses (MOOCs) provide abundant learning resources but also overwhelm learners with their sheer volume, leading to challenges such as data sparsity and cold-start issues in conventional recommendation systems. To address these challenges, we propose EGRec, a novel course recommendation model that combines knowledge graphs and Heterogeneous Graph Attention Networks to improve recommendation precision, diversity, and relevance. By integrating multimodal data, EGRec captures intricate semantic relationships between courses and knowledge points, enabling personalized and context-sensitive recommendations. Extensive experiments on real MOOCs datasets demonstrate that EGRec significantly outperforms traditional models, highlighting its potential to enhance tailored learning experiences.
- Open Access
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 70401-70413
Feature-Separated Lightweight Model for Road Extraction in Wild Environments
Extracting road information from high-resolution remote sensing images is a critical challenge, particularly in complex environments with vegetation occlusion, shadow interference, and similar terrain textures causing discontinuous or incomplete road extraction. Existing methods often suffer from high computational costs and limited accuracy. To address these issues, we propose a lightweight road extraction model based on an encoder-decoder architecture, employing CSPDarknet53 as the backbone to reduce redundant computations. A Feature Separation and Enhancement Module (FSEM) independently processes spatial and channel features, while a Feature Aggregation Module (FAM) integrates multi-scale features for improved road continuity and accuracy. This design enhances prediction performance without significantly increasing parameters, reducing computational demands and making the model more adaptable for large-scale remote sensing tasks. Experimental results demonstrate that the proposed model achieves an mIoU improvement of 1.60% and 3.15% on the DeepGlobe and Mountain Road datasets, respectively, along with over 40% runtime reduction compared to other approaches. The proposed model combines high accuracy and efficiency, offering a scalable solution for complex road extraction tasks.
- 4 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 70401-70413
Open AccessFeature-Separated Lightweight Model for Road Extraction in Wild Environments
Extracting road information from high-resolution remote sensing images is a critical challenge, particularly in complex environments with vegetation occlusion, shadow interference, and similar terrain textures causing discontinuous or incomplete road extraction. Existing methods often suffer from high computational costs and limited accuracy. To address these issues, we propose a lightweight road extraction model based on an encoder-decoder architecture, employing CSPDarknet53 as the backbone to reduce redundant computations. A Feature Separation and Enhancement Module (FSEM) independently processes spatial and channel features, while a Feature Aggregation Module (FAM) integrates multi-scale features for improved road continuity and accuracy. This design enhances prediction performance without significantly increasing parameters, reducing computational demands and making the model more adaptable for large-scale remote sensing tasks. Experimental results demonstrate that the proposed model achieves an mIoU improvement of 1.60% and 3.15% on the DeepGlobe and Mountain Road datasets, respectively, along with over 40% runtime reduction compared to other approaches. The proposed model combines high accuracy and efficiency, offering a scalable solution for complex road extraction tasks.
- 15 March 2025
- Published by Springer Nature in Electrical Engineering
- Vol. 107 (8) , 10467-10477
A new temperature prediction method for PMSM based on a nested depth neural network
- 15 March 2025
- Published by Springer Nature in Electrical Engineering
- Vol. 107 (8) , 10467-10477
A new temperature prediction method for PMSM based on a nested depth neural network
- Open Access
Asymmetric Network Combining CNN and Transformer for Building Extraction from Remote Sensing Images
The accurate extraction of buildings from remote sensing images is crucial in fields such as 3D urban planning, disaster detection, and military reconnaissance. In recent years, models based on Transformer have performed well in global information processing and contextual relationship modeling, but suffer from high computational costs and insufficient ability to capture local information. In contrast, convolutional neural networks (CNNs) are very effective in extracting local features, but have a limited ability to process global information. In this paper, an asymmetric network (CTANet), which combines the advantages of CNN and Transformer, is proposed to achieve efficient extraction of buildings. Specifically, CTANet employs ConvNeXt as an encoder to extract features and combines it with an efficient bilateral hybrid attention transformer (BHAFormer) which is designed as a decoder. The BHAFormer establishes global dependencies from both texture edge features and background information perspectives to extract buildings more accurately while maintaining a low computational cost. Additionally, the multiscale mixed attention mechanism module (MSM-AMM) is introduced to learn the multiscale semantic information and channel representations of the encoder features to reduce noise interference and compensate for the loss of information in the downsampling process. Experimental results show that the proposed model achieves the best F1-score (86.7%, 95.74%, and 90.52%) and IoU (76.52%, 91.84%, and 82.68%) compared to other state-of-the-art methods on the Massachusetts building dataset, the WHU building dataset, and the Inria aerial image labeling dataset.
Open AccessAsymmetric Network Combining CNN and Transformer for Building Extraction from Remote Sensing Images
The accurate extraction of buildings from remote sensing images is crucial in fields such as 3D urban planning, disaster detection, and military reconnaissance. In recent years, models based on Transformer have performed well in global information processing and contextual relationship modeling, but suffer from high computational costs and insufficient ability to capture local information. In contrast, convolutional neural networks (CNNs) are very effective in extracting local features, but have a limited ability to process global information. In this paper, an asymmetric network (CTANet), which combines the advantages of CNN and Transformer, is proposed to achieve efficient extraction of buildings. Specifically, CTANet employs ConvNeXt as an encoder to extract features and combines it with an efficient bilateral hybrid attention transformer (BHAFormer) which is designed as a decoder. The BHAFormer establishes global dependencies from both texture edge features and background information perspectives to extract buildings more accurately while maintaining a low computational cost. Additionally, the multiscale mixed attention mechanism module (MSM-AMM) is introduced to learn the multiscale semantic information and channel representations of the encoder features to reduce noise interference and compensate for the loss of information in the downsampling process. Experimental results show that the proposed model achieves the best F1-score (86.7%, 95.74%, and 90.52%) and IoU (76.52%, 91.84%, and 82.68%) compared to other state-of-the-art methods on the Massachusetts building dataset, the WHU building dataset, and the Inria aerial image labeling dataset.