Lei Feng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Lei Feng has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Chunyu Ning
School of Life Science and Technology, Changchun University of Science and Technology, Changchun, China
3 shared publications Mingliang Li
2 shared publications- You Tang
Electrical and Information Engineering College, JiLin Agricultural Science and Technology University, Jilin, China
2 shared publications Guanshi Ye
College of Electrical and Information Engineering, Jilin Agricultural Science and Technology University, Jilin, China
1 shared publicationLianxi Zhao
School of Life Science and Technology, Changchun University of Science and Technology, Changchun, China
1 shared publication- Mingye Li
Department of Information Systems and Business Analytics, RMIT University, Melbourne, VIC, Australia
1 shared publication
Latest Publications
- Open Access
- 24 June 2026
- Published by Elsevier in SSRN Electronic Journal
Structure-Aware Learning Framework for Animal Histopathology Analysis under Limited Annotation Conditions
Currently, animal pathology image analysis still primarily relies on manual examination, facing challenges such as high annotation costs, limited data availability, and low diagnostic efficiency. African swine fever (ASF), a highly contagious disease that severely threatens the swine industry, exhibits complex histopathological characteristics, including diverse tissue alterations and diffuse inflammatory patterns, which increase the difficulty of pathological feature modeling. To address these challenges, we propose a structure-aware learning framework for ASF histopathology analysis under limited annotation conditions. Specifically, we develop a Low-Frequency Guided Self-Supervised pretraining framework (LFG-SSL), which leverages self-distillation and a Low-Frequency Guided Transformer (LFG-Transformer) to learn pathological representations from unlabeled images. By preserving original spatial features as queries while constructing keys and values from low-frequency components, LFG-Transformer introduces globally consistent pathological priors into the attention mechanism, thereby enhancing the perception of large-scale tissue organization and inflammatory spread. For downstream classification, we further propose a Dual-Prototype Dynamic Multiple Instance Learning module (DPD-MIL), which integrates robust reference prototypes and global support prototypes to guide discriminative instance modeling. In addition, a learnable sparsity mask mechanism is introduced to suppress noisy regions. Experimental results on three ASF pathology datasets demonstrate that LFG-SSL consistently outperforms mainstream feature encoders, while DPD-MIL further improves classification performance across different MIL frameworks. Their combination achieves the best overall performance on all datasets. The proposed framework provides a promising solution for intelligent analysis of animal histopathological images.
- 24 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessStructure-Aware Learning Framework for Animal Histopathology Analysis under Limited Annotation Conditions
Currently, animal pathology image analysis still primarily relies on manual examination, facing challenges such as high annotation costs, limited data availability, and low diagnostic efficiency. African swine fever (ASF), a highly contagious disease that severely threatens the swine industry, exhibits complex histopathological characteristics, including diverse tissue alterations and diffuse inflammatory patterns, which increase the difficulty of pathological feature modeling. To address these challenges, we propose a structure-aware learning framework for ASF histopathology analysis under limited annotation conditions. Specifically, we develop a Low-Frequency Guided Self-Supervised pretraining framework (LFG-SSL), which leverages self-distillation and a Low-Frequency Guided Transformer (LFG-Transformer) to learn pathological representations from unlabeled images. By preserving original spatial features as queries while constructing keys and values from low-frequency components, LFG-Transformer introduces globally consistent pathological priors into the attention mechanism, thereby enhancing the perception of large-scale tissue organization and inflammatory spread. For downstream classification, we further propose a Dual-Prototype Dynamic Multiple Instance Learning module (DPD-MIL), which integrates robust reference prototypes and global support prototypes to guide discriminative instance modeling. In addition, a learnable sparsity mask mechanism is introduced to suppress noisy regions. Experimental results on three ASF pathology datasets demonstrate that LFG-SSL consistently outperforms mainstream feature encoders, while DPD-MIL further improves classification performance across different MIL frameworks. Their combination achieves the best overall performance on all datasets. The proposed framework provides a promising solution for intelligent analysis of animal histopathological images.
- Open Access
- 4 December 2025
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 37 (10) , 353
MCA framework: a novel multi-dimensional cooperative analysis framework for Alzheimer's disease diagnosis
Structural magnetic resonance imaging (sMRI) plays a pivotal role in the diagnosis of Alzheimer's disease (AD), owing to its high-resolution whole-brain imaging capability. Recent studies have shifted from relying on prior knowledge-based extraction of sMRI features to employing deep learning approaches for automatic feature extraction. However, the deep learning-based approaches may overlook the interrelationships among features from various dimensions, which focus solely on inter-slice information or spatial structural features, thereby limiting the comprehensiveness of brain lesion characterization. To address this limitation, we propose a multi-dimensional cooperative analysis (MCA) framework designed to integrate multi-dimensional feature information effectively. The framework comprises two stages. In the first stage, multi-frequency aware (MFA) downsampling utilizes a novel multi-frequency aware fusion module based on a three-dimensional Sym2 wavelet transform to extract multi-scale features. In the second stage, the multi-view slice learner (MVSL) and the spatial structural learner (SSL), which respectively employ a multi-view encoder and a graph hybrid attention convolutional block, collaborate to capture AD lesion features from diverse perspectives. Extensive experimental validation demonstrates that the MCA framework achieves an accuracy of 95.12% and a sensitivity of 94.07% on the ADNI dataset, outperforming recent state-of-the-art methods. It also exhibits robust generalization on the AIBL and OASIS datasets. SHAP-based analysis provides interpretability to the MCA framework by revealing the differential contributions of various brain regions in AD diagnosis.
- 4 December 2025
- Published by Springer Nature in Journal of King Saud University Computer and Information Sciences
- Vol. 37 (10) , 353
Open AccessMCA framework: a novel multi-dimensional cooperative analysis framework for Alzheimer's disease diagnosis
Structural magnetic resonance imaging (sMRI) plays a pivotal role in the diagnosis of Alzheimer's disease (AD), owing to its high-resolution whole-brain imaging capability. Recent studies have shifted from relying on prior knowledge-based extraction of sMRI features to employing deep learning approaches for automatic feature extraction. However, the deep learning-based approaches may overlook the interrelationships among features from various dimensions, which focus solely on inter-slice information or spatial structural features, thereby limiting the comprehensiveness of brain lesion characterization. To address this limitation, we propose a multi-dimensional cooperative analysis (MCA) framework designed to integrate multi-dimensional feature information effectively. The framework comprises two stages. In the first stage, multi-frequency aware (MFA) downsampling utilizes a novel multi-frequency aware fusion module based on a three-dimensional Sym2 wavelet transform to extract multi-scale features. In the second stage, the multi-view slice learner (MVSL) and the spatial structural learner (SSL), which respectively employ a multi-view encoder and a graph hybrid attention convolutional block, collaborate to capture AD lesion features from diverse perspectives. Extensive experimental validation demonstrates that the MCA framework achieves an accuracy of 95.12% and a sensitivity of 94.07% on the ADNI dataset, outperforming recent state-of-the-art methods. It also exhibits robust generalization on the AIBL and OASIS datasets. SHAP-based analysis provides interpretability to the MCA framework by revealing the differential contributions of various brain regions in AD diagnosis.
- Open Access
- 24 October 2025
- Published by Springer Nature in Plant Methods
- Vol. 21 (1) , 1-19
LDSL framework: a lightweight dual-stream learning framework for wheat disease detection
Wheat diseases significantly impair production efficiency and grain quality in the wheat industry. In recent research, deep learning techniques have been widely applied to plant disease detection. However, wheat disease images collected in field conditions often face complex backgrounds and diverse lesion shapes, making accurate disease classification difficult. In real-world applications, agricultural disease recognition systems must also deal with limited computational resources and edge device constraints, emphasizing the need for lightweight methods. To solve these challenges, this paper introduces a lightweight dual-stream learning (LDSL) framework for wheat disease detection. The framework adopts a unique global–local dual-stream architecture that combines global semantic understanding with local discriminative analysis. The global learning stream extracts comprehensive semantic features and generates saliency maps to highlight key regions, while the local learning stream performs fine-grained inspection of these regions using a novel dynamic-static dual attention (DSDA) mechanism. Additionally, a Kullback–Leibler (KL) divergence perturbation strategy is implemented during training to boost the LDSL framework’s robustness in noisy and complex settings. Experimental results show that the proposed LDSL framework achieves an accuracy of 94.44%, a precision of 94.47%, a recall of 94.44%, and an F1-score of 94.45%, outperforming several mainstream classification models in wheat disease recognition, such as ConvNeXt-T (92.66% accuracy, 92.69% precision, 92.66% recall, and 92.63% F1). The proposed LDSL framework is lightweight, using only 4.41 M parameters and 1.71G FLOPs. On the NVIDIA Jetson Orin Nano, it requires just 15.99 MB of storage, 39.49 MB of peak memory, and achieves an inference latency of 234.76 ms/image, demonstrating good potential for real-world deployment. This study provides a novel detection framework for wheat disease research, which significantly improves various classification metrics. With low parameter and computation costs, the framework demonstrates good potential for practical deployment.
- 24 October 2025
- Published by Springer Nature in Plant Methods
- Vol. 21 (1) , 1-19
Open AccessLDSL framework: a lightweight dual-stream learning framework for wheat disease detection
Wheat diseases significantly impair production efficiency and grain quality in the wheat industry. In recent research, deep learning techniques have been widely applied to plant disease detection. However, wheat disease images collected in field conditions often face complex backgrounds and diverse lesion shapes, making accurate disease classification difficult. In real-world applications, agricultural disease recognition systems must also deal with limited computational resources and edge device constraints, emphasizing the need for lightweight methods. To solve these challenges, this paper introduces a lightweight dual-stream learning (LDSL) framework for wheat disease detection. The framework adopts a unique global–local dual-stream architecture that combines global semantic understanding with local discriminative analysis. The global learning stream extracts comprehensive semantic features and generates saliency maps to highlight key regions, while the local learning stream performs fine-grained inspection of these regions using a novel dynamic-static dual attention (DSDA) mechanism. Additionally, a Kullback–Leibler (KL) divergence perturbation strategy is implemented during training to boost the LDSL framework’s robustness in noisy and complex settings. Experimental results show that the proposed LDSL framework achieves an accuracy of 94.44%, a precision of 94.47%, a recall of 94.44%, and an F1-score of 94.45%, outperforming several mainstream classification models in wheat disease recognition, such as ConvNeXt-T (92.66% accuracy, 92.69% precision, 92.66% recall, and 92.63% F1). The proposed LDSL framework is lightweight, using only 4.41 M parameters and 1.71G FLOPs. On the NVIDIA Jetson Orin Nano, it requires just 15.99 MB of storage, 39.49 MB of peak memory, and achieves an inference latency of 234.76 ms/image, demonstrating good potential for real-world deployment. This study provides a novel detection framework for wheat disease research, which significantly improves various classification metrics. With low parameter and computation costs, the framework demonstrates good potential for practical deployment.