Aokun Mei
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Aokun Mei has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Hua Huo
Henan University of Science and Technology
4 shared publications - Jiaxin Xu
College of Information Engineering and Artificial Intelligence, Henan University of Science and Technology, Luoyang 471000, China
4 shared publications - Chen Zhang
College of Information Engineering and Artificial Intelligence, Henan University of Science and Technology, Luoyang 471000, China
3 shared publications - Shilu Kang
School of Information Engineering, Henan University of Science and Technology, Luoyang 471023, China
2 shared publications - Lan Ma
School of Information Engineering, Henan University of Science and Technology, Luoyang 471023, China
1 shared publication - Yinpeng Guo
School of Information Engineering, Henan University of Science and Technology, Luoyang 471023, China
1 shared publication
Latest Publications
- Open Access
BGD-Net: An Object–Background Decoupling Network for Oriented Object Detection in Complex Remote Sensing Scenes
Oriented object detection in high-resolution remote sensing images remains vulnerable to object–background confusion in complex scenes, where visually similar background structures can produce high-confidence false positives. To address this problem, we propose BGD-Net, an object–background decoupling framework that suppresses confusing background responses from the feature, sample, and optimization levels. First, a Background-Decoupled Feature Module (BDFM) separately models object responses and background activations and performs residual feature purification before proposal generation. Second, Hard Background Mining (HBM) identifies high-confidence, low-overlap background RoIs that are most likely to cause false-positive detections and assigns them greater training emphasis. Third, an Object–Background Contrastive Suppression Loss (OBCS Loss) uses the mined hard backgrounds as targeted negative samples to enlarge the representation gap between true objects and confusing backgrounds in the RoI embedding space. Experiments on DOTA-v1.0 show that BGD-Net achieves 82.94% mAP@0.5, improving the Oriented R-CNN baseline by 2.07 percentage points while reducing Hard-FP by 38.48%. The method also achieves 98.47% mAP@0.5 on HRSC2016 and 68.35% mAP@0.5 on DIOR-R, demonstrating effective performance across different remote sensing scenes. Ablation, multi-seed, and threshold-sensitivity experiments further verify the complementary contributions and reproducibility of the proposed components. BGD-Net introduces only a modest increase in model complexity, indicating a favorable balance between detection accuracy and practical efficiency.
Open AccessBGD-Net: An Object–Background Decoupling Network for Oriented Object Detection in Complex Remote Sensing Scenes
Oriented object detection in high-resolution remote sensing images remains vulnerable to object–background confusion in complex scenes, where visually similar background structures can produce high-confidence false positives. To address this problem, we propose BGD-Net, an object–background decoupling framework that suppresses confusing background responses from the feature, sample, and optimization levels. First, a Background-Decoupled Feature Module (BDFM) separately models object responses and background activations and performs residual feature purification before proposal generation. Second, Hard Background Mining (HBM) identifies high-confidence, low-overlap background RoIs that are most likely to cause false-positive detections and assigns them greater training emphasis. Third, an Object–Background Contrastive Suppression Loss (OBCS Loss) uses the mined hard backgrounds as targeted negative samples to enlarge the representation gap between true objects and confusing backgrounds in the RoI embedding space. Experiments on DOTA-v1.0 show that BGD-Net achieves 82.94% mAP@0.5, improving the Oriented R-CNN baseline by 2.07 percentage points while reducing Hard-FP by 38.48%. The method also achieves 98.47% mAP@0.5 on HRSC2016 and 68.35% mAP@0.5 on DIOR-R, demonstrating effective performance across different remote sensing scenes. Ablation, multi-seed, and threshold-sensitivity experiments further verify the complementary contributions and reproducibility of the proposed components. BGD-Net introduces only a modest increase in model complexity, indicating a favorable balance between detection accuracy and practical efficiency.
- Open Access
- 26 June 2026
- Published by MDPI AG in Electronics
- Vol. 15 (13) , 2819
OPT-Net: An Orientation-Preserving Transformer for End-to-End Oriented Object Detection in Remote Sensing Images
The objects in high-resolution remote sensing images usually exhibit arbitrary orientations, multi-scale variations, dense distributions, and complex background interference, posing significant challenges to oriented object detection. Although existing DETR-style end-to-end detectors eliminate the need for anchor design and non-maximum suppression, they still suffer from insufficient orientation priors in object queries, limited orientation consistency in decoder feature interaction, and unstable set matching for oriented bounding boxes. To address these issues, this paper proposes an end-to-end Transformer framework, termed OPT-Net (Orientation-Preserving Transformer Network), for oriented object detection in remote sensing images. OPT-Net treats orientation information as a structured geometric prior and propagates it through query initialization, feature interaction, and matching optimization. Specifically, an Orientation-Aware Query Initialization (OAQI) module is designed to generate initial queries using center confidence and orientation priors. An Orientation-Consistent Cross-Attention (OCCA) mechanism is proposed to perform orientation-conditioned modulation on Value features while keeping the standard Query–Key attention computation unchanged. Furthermore, an Uncertainty-aware Matching Loss (UML) is introduced to incorporate instance-level geometric uncertainty into Hungarian matching and regression optimization. Experimental results on the DOTA-v1.0 and HRSC2016 datasets show that OPT-Net achieves 76.83% and 90.58% mAP, respectively, demonstrating competitive detection accuracy and adaptability to complex remote sensing scenarios. Ablation studies and visualization results further validate the effectiveness of each proposed module.
- 26 June 2026
- Published by MDPI AG in Electronics
- Vol. 15 (13) , 2819
Open AccessOPT-Net: An Orientation-Preserving Transformer for End-to-End Oriented Object Detection in Remote Sensing Images
The objects in high-resolution remote sensing images usually exhibit arbitrary orientations, multi-scale variations, dense distributions, and complex background interference, posing significant challenges to oriented object detection. Although existing DETR-style end-to-end detectors eliminate the need for anchor design and non-maximum suppression, they still suffer from insufficient orientation priors in object queries, limited orientation consistency in decoder feature interaction, and unstable set matching for oriented bounding boxes. To address these issues, this paper proposes an end-to-end Transformer framework, termed OPT-Net (Orientation-Preserving Transformer Network), for oriented object detection in remote sensing images. OPT-Net treats orientation information as a structured geometric prior and propagates it through query initialization, feature interaction, and matching optimization. Specifically, an Orientation-Aware Query Initialization (OAQI) module is designed to generate initial queries using center confidence and orientation priors. An Orientation-Consistent Cross-Attention (OCCA) mechanism is proposed to perform orientation-conditioned modulation on Value features while keeping the standard Query–Key attention computation unchanged. Furthermore, an Uncertainty-aware Matching Loss (UML) is introduced to incorporate instance-level geometric uncertainty into Hungarian matching and regression optimization. Experimental results on the DOTA-v1.0 and HRSC2016 datasets show that OPT-Net achieves 76.83% and 90.58% mAP, respectively, demonstrating competitive detection accuracy and adaptability to complex remote sensing scenarios. Ablation studies and visualization results further validate the effectiveness of each proposed module.
- Open Access
- 30 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (9) , 1904
A Multimodal Audiovisual Deep Learning Framework for Early Detection of Parkinson’s Disease
Parkinson’s disease (PD) is a progressive neurodegenerative disorder primarily caused by the degeneration of dopamine-producing neurons in the substantia nigra, leading to characteristic motor symptoms such as tremors, rigidity, and bradykinesia, as well as non-motor manifestations including depression, sleep disturbances, and speech impairments. Among these symptoms, speech abnormalities affect approximately 90% of individuals with PD, making acoustic analysis a promising non-invasive cue for early detection. However, subtle speech variations are often imperceptible to the human ear, and speech-only analysis may overlook complementary visual manifestations, such as hypomimia—reduced facial expressivity commonly observed in PD patients. To address these limitations, we propose Parkinson’s Detection via Attentional Fusion Network (PDAF-Net), a novel multimodal deep learning framework for early PD detection that jointly models acoustic and facial dynamic features in a binary classification setting. The proposed architecture consists of a Dual-Stream Feature Encoder (DSFE), with an audio branch based on a one-dimensional convolutional neural network (1D-CNN) and bidirectional long short-term memory (BiLSTM), and a visual branch built upon a two-dimensional convolutional neural network (2D-CNN) and a Transformer encoder. Multimodal integration is achieved through a Cross-Attention-guided Attentional Feature Fusion (CA-AFF) module, which explicitly models bidirectional cross-modal interactions and performs adaptive feature recalibration via an iterative attentional fusion mechanism. We conducted experiments on a self-collected Chinese multimodal dataset comprising 100 PD patients and 100 healthy controls. Although the data are balanced at the subject level, sliding-window segmentation introduces sample-level imbalance; to address this issue, a class-balanced focal loss is employed. Model performance was evaluated using subject-wise five-fold cross-validation. The results demonstrate that PDAF-Net consistently outperforms unimodal baselines across multiple evaluation metrics, achieving an accuracy of 89.3%, an F1-score of 0.884, and an AUC of 0.916. These findings highlight the effectiveness of explicit cross-modal interaction modeling and adaptive feature fusion for improving automated early PD screening in real-world clinical settings.
- 30 April 2026
- Published by MDPI AG in Electronics
- Vol. 15 (9) , 1904
Open AccessA Multimodal Audiovisual Deep Learning Framework for Early Detection of Parkinson’s Disease
Parkinson’s disease (PD) is a progressive neurodegenerative disorder primarily caused by the degeneration of dopamine-producing neurons in the substantia nigra, leading to characteristic motor symptoms such as tremors, rigidity, and bradykinesia, as well as non-motor manifestations including depression, sleep disturbances, and speech impairments. Among these symptoms, speech abnormalities affect approximately 90% of individuals with PD, making acoustic analysis a promising non-invasive cue for early detection. However, subtle speech variations are often imperceptible to the human ear, and speech-only analysis may overlook complementary visual manifestations, such as hypomimia—reduced facial expressivity commonly observed in PD patients. To address these limitations, we propose Parkinson’s Detection via Attentional Fusion Network (PDAF-Net), a novel multimodal deep learning framework for early PD detection that jointly models acoustic and facial dynamic features in a binary classification setting. The proposed architecture consists of a Dual-Stream Feature Encoder (DSFE), with an audio branch based on a one-dimensional convolutional neural network (1D-CNN) and bidirectional long short-term memory (BiLSTM), and a visual branch built upon a two-dimensional convolutional neural network (2D-CNN) and a Transformer encoder. Multimodal integration is achieved through a Cross-Attention-guided Attentional Feature Fusion (CA-AFF) module, which explicitly models bidirectional cross-modal interactions and performs adaptive feature recalibration via an iterative attentional fusion mechanism. We conducted experiments on a self-collected Chinese multimodal dataset comprising 100 PD patients and 100 healthy controls. Although the data are balanced at the subject level, sliding-window segmentation introduces sample-level imbalance; to address this issue, a class-balanced focal loss is employed. Model performance was evaluated using subject-wise five-fold cross-validation. The results demonstrate that PDAF-Net consistently outperforms unimodal baselines across multiple evaluation metrics, achieving an accuracy of 89.3%, an F1-score of 0.884, and an AUC of 0.916. These findings highlight the effectiveness of explicit cross-modal interaction modeling and adaptive feature fusion for improving automated early PD screening in real-world clinical settings.
- Open Access
Rotation-Sensitive Feature Enhancement Network for Oriented Object Detection in Remote Sensing Images
Oriented object detection in remote sensing images remains a challenging task due to arbitrary target rotations, extreme scale variations, and complex backgrounds. However, current rotated detectors still face several limitations: insufficient orientation-sensitive feature representation, feature misalignment for rotated proposals, and unstable optimization of rotation parameters. To address these issues, this paper proposes an enhanced Rotation-Sensitive Feature Pyramid Network (RSFPN) framework. Building upon the effective Oriented R-CNN paradigm, we introduce three novel core components: (1) a Dynamic Adaptive Feature Pyramid Network (DAFPN) that enables bidirectional multi-scale feature fusion through semantic-guided upsampling and structure-enhanced downsampling paths; (2) an Angle-Aware Collaborative Attention (AACA) module that incorporates orientation priors to guide feature refinement; (3) a Geometrically Consistent Multi-Task Loss (GC-MTL) that unifies the regression of rotation parameters with periodic smoothing and adaptive weight mechanisms. Comprehensive experiments on the DOTA-v1.0 and HRSC2016 benchmarks show that our RSFPN achieves superior performance. It attains a state-of-the-art mAP of 77.42% on DOTA-v1.0 and 91.85% on HRSC2016, while maintaining efficient inference at 14.5 FPS, demonstrating a favorable accuracy-efficiency trade-off. Visual analysis confirms that our method produces concentrated, rotation-aware feature responses and effectively suppresses background interference. The proposed approach provides a robust solution for detecting multi-oriented objects in high-resolution remote sensing imagery, with significant practical value for urban planning, environmental monitoring, and security applications.
Open AccessRotation-Sensitive Feature Enhancement Network for Oriented Object Detection in Remote Sensing Images
Oriented object detection in remote sensing images remains a challenging task due to arbitrary target rotations, extreme scale variations, and complex backgrounds. However, current rotated detectors still face several limitations: insufficient orientation-sensitive feature representation, feature misalignment for rotated proposals, and unstable optimization of rotation parameters. To address these issues, this paper proposes an enhanced Rotation-Sensitive Feature Pyramid Network (RSFPN) framework. Building upon the effective Oriented R-CNN paradigm, we introduce three novel core components: (1) a Dynamic Adaptive Feature Pyramid Network (DAFPN) that enables bidirectional multi-scale feature fusion through semantic-guided upsampling and structure-enhanced downsampling paths; (2) an Angle-Aware Collaborative Attention (AACA) module that incorporates orientation priors to guide feature refinement; (3) a Geometrically Consistent Multi-Task Loss (GC-MTL) that unifies the regression of rotation parameters with periodic smoothing and adaptive weight mechanisms. Comprehensive experiments on the DOTA-v1.0 and HRSC2016 benchmarks show that our RSFPN achieves superior performance. It attains a state-of-the-art mAP of 77.42% on DOTA-v1.0 and 91.85% on HRSC2016, while maintaining efficient inference at 14.5 FPS, demonstrating a favorable accuracy-efficiency trade-off. Visual analysis confirms that our method produces concentrated, rotation-aware feature responses and effectively suppresses background interference. The proposed approach provides a robust solution for detecting multi-oriented objects in high-resolution remote sensing imagery, with significant practical value for urban planning, environmental monitoring, and security applications.