Mengya Han
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
Mengya Han has a total of 48 co-authors. Here are the top co-authors by number of shared publications:
- Yong Luo
Wuhan University
6 shared publications - Bo Du
School of Computer Science, Wuhan University, Wuhan, China
4 shared publications - Han Hu
Beijing Institute of Technology
4 shared publications - Dacheng Tao
Nanyang Technological University
3 shared publications - Yibing Zhan
Wuhan University
3 shared publications - Yonggang Wen
Nanyang technological university, Singapore, Singapore
2 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 10548-10557
DVLA: A Dynamic Vision-Language Alignment Framework for Mitigating Hallucination in LVLMs
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 10548-10557
Open AccessDVLA: A Dynamic Vision-Language Alignment Framework for Mitigating Hallucination in LVLMs
- 13 July 2026
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 37 (28) , 285402
Frequency-domain heterogeneous differential spatial-aware network for transmission line anomalous object detection
Recently, deep neural networks have achieved remarkable progress in anomalous object detection for power transmission lines. However, existing object detection methods inadequately leverage the complementary characteristics of features from different frequency components, leading to the loss of both global structures and local details under complex inspection scenarios. Moreover, most cross-level feature fusion mechanisms still fail to effectively capture spatial dependencies, limiting spatial awareness and contextual representation for small-scale anomalous targets. To address these issues, we propose a frequency-domain heterogeneous differential spatial-aware network (FHDS-Net) for transmission line anomalous object detection. Specifically, we propose a frequency-domain heterogeneous expert modulation module that integrates bidirectional complementary pathways and high- and low-frequency experts to exploit high- and low-frequency complementarity and capture long-range dependencies via multi-directional scanning, thereby improving the joint representation of global structures and local details. Furthermore, the differential spatial-aware alignment fusion module employs a two-stage progressive process for coarse-to-fine cross-level alignment, where window-level coarse association anchors structurally relevant regions and pixel-level fine attention recalibrates localized responses, thereby facilitating a bidirectional synergy between shallow geometric textures and deep semantic context to effectively enhance the discriminative representation of subtle, small-scale anomalies. Experimental results on two challenging benchmark datasets demonstrate the superiority of our FHDS-Net over several state-of-the-art methods for anomalous object detection. The code is released publicly at https://github.com/h-519/FHDS-Net .
- 13 July 2026
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 37 (28) , 285402
Frequency-domain heterogeneous differential spatial-aware network for transmission line anomalous object detection
Recently, deep neural networks have achieved remarkable progress in anomalous object detection for power transmission lines. However, existing object detection methods inadequately leverage the complementary characteristics of features from different frequency components, leading to the loss of both global structures and local details under complex inspection scenarios. Moreover, most cross-level feature fusion mechanisms still fail to effectively capture spatial dependencies, limiting spatial awareness and contextual representation for small-scale anomalous targets. To address these issues, we propose a frequency-domain heterogeneous differential spatial-aware network (FHDS-Net) for transmission line anomalous object detection. Specifically, we propose a frequency-domain heterogeneous expert modulation module that integrates bidirectional complementary pathways and high- and low-frequency experts to exploit high- and low-frequency complementarity and capture long-range dependencies via multi-directional scanning, thereby improving the joint representation of global structures and local details. Furthermore, the differential spatial-aware alignment fusion module employs a two-stage progressive process for coarse-to-fine cross-level alignment, where window-level coarse association anchors structurally relevant regions and pixel-level fine attention recalibrates localized responses, thereby facilitating a bidirectional synergy between shallow geometric textures and deep semantic context to effectively enhance the discriminative representation of subtle, small-scale anomalies. Experimental results on two challenging benchmark datasets demonstrate the superiority of our FHDS-Net over several state-of-the-art methods for anomalous object detection. The code is released publicly at https://github.com/h-519/FHDS-Net .
- Open Access
- 25 February 2026
- Published by Elsevier in SSRN Electronic Journal
BiCPF: Bi-Conditional Parallel-Fusion Network for Composed Image Retrieval
Composed Image Retrieval (CIR) aims to retrieve a target image given a reference image and a modification text. However, existing fusion-based methods often produce modality dominant queries that consistently lean toward a certain modality, thereby impairing the model's ability to effectively handle diverse real-world scenarios. In contrast, humans performing composed image retrieval dynamically reinterpret visual content from the perspective of text while simultaneously grounding language understanding in visual context, forming a bidirectional cross-modal semantic construction process. Motivated by this, we propose a Bi-Conditional Parallel-Fusion Network (\textit{BiCPF}) that models cross-modal semantic relations from two complementary perspectives: an image-conditioned branch interprets textual modifications within the visual context, while a text-conditioned branch reinterprets visual semantics guided by textual cues. Specifically, our \textit{BiCPF} includes: 1) Bi-Conditional Parallel-Fusion Module that utilizes two parallel fusion layers to learn modality-grounded queries conditioned on the complementary modality; 2) Similarity Aggregation Module that dynamically balances the contributions of textual and visual similarities for adaptive retrieval to handle diverse retrieval scenarios. Extensive experiments conducted on three benchmarks demonstrate that our method outperforms existing state-of-the-art approaches, particularly in scenarios where treating one modality as dominance is insufficient for accurate retrieval.
- 25 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessBiCPF: Bi-Conditional Parallel-Fusion Network for Composed Image Retrieval
Composed Image Retrieval (CIR) aims to retrieve a target image given a reference image and a modification text. However, existing fusion-based methods often produce modality dominant queries that consistently lean toward a certain modality, thereby impairing the model's ability to effectively handle diverse real-world scenarios. In contrast, humans performing composed image retrieval dynamically reinterpret visual content from the perspective of text while simultaneously grounding language understanding in visual context, forming a bidirectional cross-modal semantic construction process. Motivated by this, we propose a Bi-Conditional Parallel-Fusion Network (\textit{BiCPF}) that models cross-modal semantic relations from two complementary perspectives: an image-conditioned branch interprets textual modifications within the visual context, while a text-conditioned branch reinterprets visual semantics guided by textual cues. Specifically, our \textit{BiCPF} includes: 1) Bi-Conditional Parallel-Fusion Module that utilizes two parallel fusion layers to learn modality-grounded queries conditioned on the complementary modality; 2) Similarity Aggregation Module that dynamically balances the contributions of textual and visual similarities for adaptive retrieval to handle diverse retrieval scenarios. Extensive experiments conducted on three benchmarks demonstrate that our method outperforms existing state-of-the-art approaches, particularly in scenarios where treating one modality as dominance is insufficient for accurate retrieval.
- 23 January 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 337
OACI: Object-aware contextual integration for image captioning
- 23 January 2026
- Published by Elsevier in Knowledge-Based Systems
- Vol. 337
OACI: Object-aware contextual integration for image captioning
- 4 August 2025
- Published by Springer Nature in International Journal of Computer Vision
- Vol. 133 (11) , 7553-7569
DM-PCL: Text-Driven Dual-Modal Prototype Consistency Learning for Weakly-Supervised Few-Shot Part Segmentation
- 4 August 2025
- Published by Springer Nature in International Journal of Computer Vision
- Vol. 133 (11) , 7553-7569
DM-PCL: Text-Driven Dual-Modal Prototype Consistency Learning for Weakly-Supervised Few-Shot Part Segmentation
- 4 January 2025
- Published by Elsevier in Pattern Recognition
- Vol. 162
PartSeg: Few-shot part segmentation via part-aware prompt learning
- 4 January 2025
- Published by Elsevier in Pattern Recognition
- Vol. 162
PartSeg: Few-shot part segmentation via part-aware prompt learning
- 14 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 2408-2418
Textual Enhanced Adaptive Meta-Fusion for Few-Shot Visual Recognition
Few-shot learning (FSL) is a challenging task that aims to train a classifier to recognize novel categories, where only a few annotated examples are available in each category. Recently, many FSL approaches have been proposed based on the meta-learning paradigm, which attempts to learn transferable knowledge from similar tasks by designing a meta-learner. However, most of these approaches only exploit the information from visual modality and do not utilize ones from additional modalities ( e.g. , textual description). Since the labeled examples in FSL are limited, increasing the information on the examples is a probable solution to improve the classification performance. This motivates us to propose a novel meta-learning method, termed textual enhanced adaptive meta-fusion FSL (TAMF-FSL), which leverages both the visual information from the visual image and semantic information from language supervision. Specifically, TAMF-FSL exploits the semantic information of textual description to improve the visual-based models. We first employ a text encoder to learn the semantic features of each visual category, and then design a modality alignment module and meta-fusion module to align and fuse the visual and semantic features for final prediction. Extensive experiments show that the proposed method outperforms many recent or competitive FSL counterparts on two popular datasets.
- 14 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 26 (15209210) , 2408-2418
Textual Enhanced Adaptive Meta-Fusion for Few-Shot Visual Recognition
Few-shot learning (FSL) is a challenging task that aims to train a classifier to recognize novel categories, where only a few annotated examples are available in each category. Recently, many FSL approaches have been proposed based on the meta-learning paradigm, which attempts to learn transferable knowledge from similar tasks by designing a meta-learner. However, most of these approaches only exploit the information from visual modality and do not utilize ones from additional modalities ( e.g. , textual description). Since the labeled examples in FSL are limited, increasing the information on the examples is a probable solution to improve the classification performance. This motivates us to propose a novel meta-learning method, termed textual enhanced adaptive meta-fusion FSL (TAMF-FSL), which leverages both the visual information from the visual image and semantic information from language supervision. Specifically, TAMF-FSL exploits the semantic information of textual description to improve the visual-based models. We first employ a text encoder to learn the semantic features of each visual category, and then design a modality alignment module and meta-fusion module to align and fuse the visual and semantic features for final prediction. Extensive experiments show that the proposed method outperforms many recent or competitive FSL counterparts on two popular datasets.
- 2 March 2023
- Published by Springer Nature in Machine Intelligence Research
- Vol. 20 (4) , 554-568
Region-adaptive Concept Aggregation for Few-shot Visual Recognition
- 2 March 2023
- Published by Springer Nature in Machine Intelligence Research
- Vol. 20 (4) , 554-568
Region-adaptive Concept Aggregation for Few-shot Visual Recognition
- 22 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 35 (4) , 5447-5460
Not All Instances Contribute Equally: Instance-Adaptive Class Representation Learning for Few-Shot Visual Recognition
Few-shot visual recognition refers to recognize novel visual concepts from a few labeled instances. Many few-shot visual recognition methods adopt the metric-based meta-learning paradigm by comparing the query representation with class representations to predict the category of query instance. However, the current metric-based methods generally treat all instances equally and consequently often obtain biased class representation, considering not all instances are equally significant when summarizing the instance-level representations for the class-level representation. For example, some instances may contain unrepresentative information, such as too much background and information of unrelated concepts, which skew the results. To address the above issues, we propose a novel metric-based meta-learning framework termed instance-adaptive class representation learning network (ICRL-Net) for few-shot visual recognition. Specifically, we develop an adaptive instance revaluing network (AIRN) with the capability to address the biased representation issue when generating the class representation, by learning and assigning adaptive weights for different instances according to their relative significance in the support set of corresponding class. In addition, we design an improved bilinear instance representation and incorporate two novel structural losses, i.e., intraclass instance clustering loss and interclass representation distinguishing loss, to further regulate the instance revaluation process and refine the class representation. We conduct extensive experiments on four commonly adopted few-shot benchmarks: miniImageNet, tieredImageNet, CIFAR-FS, and FC100 datasets. The experimental results compared with the state-of-the-art approaches demonstrate the superiority of our ICRL-Net.
- 22 September 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 35 (4) , 5447-5460
Not All Instances Contribute Equally: Instance-Adaptive Class Representation Learning for Few-Shot Visual Recognition
Few-shot visual recognition refers to recognize novel visual concepts from a few labeled instances. Many few-shot visual recognition methods adopt the metric-based meta-learning paradigm by comparing the query representation with class representations to predict the category of query instance. However, the current metric-based methods generally treat all instances equally and consequently often obtain biased class representation, considering not all instances are equally significant when summarizing the instance-level representations for the class-level representation. For example, some instances may contain unrepresentative information, such as too much background and information of unrelated concepts, which skew the results. To address the above issues, we propose a novel metric-based meta-learning framework termed instance-adaptive class representation learning network (ICRL-Net) for few-shot visual recognition. Specifically, we develop an adaptive instance revaluing network (AIRN) with the capability to address the biased representation issue when generating the class representation, by learning and assigning adaptive weights for different instances according to their relative significance in the support set of corresponding class. In addition, we design an improved bilinear instance representation and incorporate two novel structural losses, i.e., intraclass instance clustering loss and interclass representation distinguishing loss, to further regulate the instance revaluation process and refine the class representation. We conduct extensive experiments on four commonly adopted few-shot benchmarks: miniImageNet, tieredImageNet, CIFAR-FS, and FC100 datasets. The experimental results compared with the state-of-the-art approaches demonstrate the superiority of our ICRL-Net.
- 1 March 2019
- Published by Elsevier in Tetrahedron
- Vol. 75 (10) , 1450-1456
Palladium-catalyzed tandem cyclization of fluorinated imidoyl chlorides with 2-bromophenylboronic acid: Synthesis of 6-fluoroalkyl-phenanthridines
- 1 March 2019
- Published by Elsevier in Tetrahedron
- Vol. 75 (10) , 1450-1456
Palladium-catalyzed tandem cyclization of fluorinated imidoyl chlorides with 2-bromophenylboronic acid: Synthesis of 6-fluoroalkyl-phenanthridines