Yanan Zhu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yanan Zhu has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Jiaqiu Ai
School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, China
5 shared publications - Weibao Xue
School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, China
4 shared publications - Gui Gao
Southwest Jiaotong University
2 shared publications - Xinyu Sun
School of Computer Science and Information Engineering, Hefei University of Technology, Hefei, China
2 shared publications - Chengbo Wang
Department of Automation, School of Information Science and Technology, University of Science and Technology of China, Hefei, China
1 shared publication - Dan Guo
Hefei University of Technology, Hefei, China
1 shared publication
Latest Publications
- Top Cited Papers
- 12 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 62 (00189251) , 6547-6560
LMCNet: Lightweight Modality Compensation Network Via Knowledge Distillation for Salient Ship Detection Under Missing-Modality Conditions
Salient ship detection is critical for maritime applications that require accurate localization. Although the fusion of Synthetic Aperture Radar (SAR) and Automatic Identification System (AIS) data has proven effective in enhancing saliency and suppressing background clutter, practical deployment faces two major limitations: the limited unavailability of AIS data and the high computational overhead of existing multimodal models. These limitations pose a fundamental challenge in balancing detection accuracy and efficiency under missing modality and resource-constrained environments. To address this issue, a Lightweight Modality Compensation Network (LMCNet) is proposed. A multimodal teacher network is trained with SAR and AIS inputs to learn rich and complementary representations. Meanwhile, a compact, single-modality student network that relies only on SAR is designed to support low-cost, real-time deployment. To enable robust knowledge compensation and transfer, this paper designs a knowledge distillation strategy consisting of three modules: structure-aware attention distillation for spatial alignment, cross-head teacher distillation for semantic enhancement, and adaptive loss scheduling for dynamic optimization. This unified design allows the student model to inherit spatial precision and semantic awareness from the teacher, achieving strong performance even with limited input modalities. Extensive experiments on two datasets show that our distilled student model improves Eξ E by 1.39% and Fwβ by 4.37% compared to state-of-the-art methods, demonstrating its superior balance of accuracy and efficiency in real-world deployment scenarios.
- 12 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Aerospace and Electronic Systems
- Vol. 62 (00189251) , 6547-6560
Top Cited PapersLMCNet: Lightweight Modality Compensation Network Via Knowledge Distillation for Salient Ship Detection Under Missing-Modality Conditions
Salient ship detection is critical for maritime applications that require accurate localization. Although the fusion of Synthetic Aperture Radar (SAR) and Automatic Identification System (AIS) data has proven effective in enhancing saliency and suppressing background clutter, practical deployment faces two major limitations: the limited unavailability of AIS data and the high computational overhead of existing multimodal models. These limitations pose a fundamental challenge in balancing detection accuracy and efficiency under missing modality and resource-constrained environments. To address this issue, a Lightweight Modality Compensation Network (LMCNet) is proposed. A multimodal teacher network is trained with SAR and AIS inputs to learn rich and complementary representations. Meanwhile, a compact, single-modality student network that relies only on SAR is designed to support low-cost, real-time deployment. To enable robust knowledge compensation and transfer, this paper designs a knowledge distillation strategy consisting of three modules: structure-aware attention distillation for spatial alignment, cross-head teacher distillation for semantic enhancement, and adaptive loss scheduling for dynamic optimization. This unified design allows the student model to inherit spatial precision and semantic awareness from the teacher, achieving strong performance even with limited input modalities. Extensive experiments on two datasets show that our distilled student model improves Eξ E by 1.39% and Fwβ by 4.37% compared to state-of-the-art methods, demonstrating its superior balance of accuracy and efficiency in real-world deployment scenarios.
- 25 November 2025
- Published by Elsevier in Ocean Engineering
- Vol. 344
SFCMDA:Source-free cross-modal domain adaptation for infrared-to-SAR ship detection
- 25 November 2025
- Published by Elsevier in Ocean Engineering
- Vol. 344
SFCMDA:Source-free cross-modal domain adaptation for infrared-to-SAR ship detection
- 20 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 22 (1545598X) , 1-5
MU-Net: An Efficient Small Ship Detection Network Based on Multimodal Fusion With Unpaired SAR/AIS Data
Ship detection in Synthetic Aperture Radar (SAR) imagery has achieved significant progress but still faces several challenges. For example, single-modal SAR provides limited ship information, and identifying small ships in complex environments is difficult, affecting detection robustness. To address these issues, this letter proposes an efficient small ship detection network based on multi-modal fusion with unpaired SAR and AIS data, named MU-Net. To address the limitations of insufficient single-modal features, it is the first attempt to integrate SAR imagery with ship distribution density information from long-term AIS. Specifically, an Asymmetric Fusion Module (AFM) is designed to incorporate AIS spatial information into SAR imagery through weighted fusion, effectively overcoming the strict data pairing requirements of traditional multi-modal fusion methods and enhancing the spatial and texture feature representation of ships. Furthermore, to boost the performance of small ship detection in challenging scenarios, MU-Net designs a Feature Dynamic Optimization Module (FDOM) and a Channel Shuffle Module (CSM). FDOM adaptively adjusts convolutional filters, and CSM performs channel reconstruction, enhancing the multi-modal feature representation of small ships and effectively reducing interference from complex backgrounds. With the integration of multi-modal information, MU-Net achieves higher detection accuracy using a single detection head. Experimental results indicate that the proposed method outperforms state-of-the-art models, achieving 89.81% Precision, 90.47% Recall, 94.82% AP0.5, and 54.75 AP0.5:0.95, with only 5.25M parameters and 55.9 GFLOPs. The dataset and code will be available at: https://github.com/DLRS09/MU-Net.
- 20 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Geoscience and Remote Sensing Letters
- Vol. 22 (1545598X) , 1-5
MU-Net: An Efficient Small Ship Detection Network Based on Multimodal Fusion With Unpaired SAR/AIS Data
Ship detection in Synthetic Aperture Radar (SAR) imagery has achieved significant progress but still faces several challenges. For example, single-modal SAR provides limited ship information, and identifying small ships in complex environments is difficult, affecting detection robustness. To address these issues, this letter proposes an efficient small ship detection network based on multi-modal fusion with unpaired SAR and AIS data, named MU-Net. To address the limitations of insufficient single-modal features, it is the first attempt to integrate SAR imagery with ship distribution density information from long-term AIS. Specifically, an Asymmetric Fusion Module (AFM) is designed to incorporate AIS spatial information into SAR imagery through weighted fusion, effectively overcoming the strict data pairing requirements of traditional multi-modal fusion methods and enhancing the spatial and texture feature representation of ships. Furthermore, to boost the performance of small ship detection in challenging scenarios, MU-Net designs a Feature Dynamic Optimization Module (FDOM) and a Channel Shuffle Module (CSM). FDOM adaptively adjusts convolutional filters, and CSM performs channel reconstruction, enhancing the multi-modal feature representation of small ships and effectively reducing interference from complex backgrounds. With the integration of multi-modal information, MU-Net achieves higher detection accuracy using a single detection head. Experimental results indicate that the proposed method outperforms state-of-the-art models, achieving 89.81% Precision, 90.47% Recall, 94.82% AP0.5, and 54.75 AP0.5:0.95, with only 5.25M parameters and 55.9 GFLOPs. The dataset and code will be available at: https://github.com/DLRS09/MU-Net.
- 27 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (8) , 13344-13356
Cross-Modal Ship Detection From Optical to SAR Images Based on Pixel- and Feature-Level Progressive Transfer
Synthetic Aperture Radar (SAR) ship detection has advanced through deep learning but often requires many accurately annotated SAR images, which are time-consuming and labor-intensive. To address this, we propose an unsupervised cross-modal adaptation method based on pixel-level and feature-level progressive transfer (CD-PFPT). CD-PFPT effectively transfers knowledge from optical to SAR domains, enhancing ship detection performance in unlabeled SAR images. Our approach consists of two components: pixel-level and feature-level cross-domain adaptation. At the pixel level, a texture-enhanced style generation network called Edge CycleGAN is designed. This network integrates the Haar wavelet transform to convert spatial into frequency domain information, ensuring that the generated images emphasize more on the texture and structural information, while effectively reducing common artifacts and distortions in cross-modal image translation. Additionally, the proposed novel edge loss function ensures that Edge CycleGAN generates samples matching the SAR image distribution. At the feature level, we propose a Multi-Scale Feature Enhancement and Denoising (MFED) module, which uses spatial attention to suppress background noise and multi-scale convolutional extraction to capture diverse ship sizes. Furthermore, a cross-domain loss is proposed to calculate the mean embedding distance in the Reproducing Kernel Hilbert Space (RKHS) using multiple kernel functions, thereby reducing distribution differences between optical and SAR images. In cross-domain tasks from the optical dataset of HRRSD to the SAR dataset of SAR-Ship, our method improves the F1-score by up to 3.9% and the mAP0.5 by up to 6.1% compared to the next best semi-supervised method.
- 27 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (8) , 13344-13356
Cross-Modal Ship Detection From Optical to SAR Images Based on Pixel- and Feature-Level Progressive Transfer
Synthetic Aperture Radar (SAR) ship detection has advanced through deep learning but often requires many accurately annotated SAR images, which are time-consuming and labor-intensive. To address this, we propose an unsupervised cross-modal adaptation method based on pixel-level and feature-level progressive transfer (CD-PFPT). CD-PFPT effectively transfers knowledge from optical to SAR domains, enhancing ship detection performance in unlabeled SAR images. Our approach consists of two components: pixel-level and feature-level cross-domain adaptation. At the pixel level, a texture-enhanced style generation network called Edge CycleGAN is designed. This network integrates the Haar wavelet transform to convert spatial into frequency domain information, ensuring that the generated images emphasize more on the texture and structural information, while effectively reducing common artifacts and distortions in cross-modal image translation. Additionally, the proposed novel edge loss function ensures that Edge CycleGAN generates samples matching the SAR image distribution. At the feature level, we propose a Multi-Scale Feature Enhancement and Denoising (MFED) module, which uses spatial attention to suppress background noise and multi-scale convolutional extraction to capture diverse ship sizes. Furthermore, a cross-domain loss is proposed to calculate the mean embedding distance in the Reproducing Kernel Hilbert Space (RKHS) using multiple kernel functions, thereby reducing distribution differences between optical and SAR images. In cross-domain tasks from the optical dataset of HRRSD to the SAR dataset of SAR-Ship, our method improves the F1-score by up to 3.9% and the mAP0.5 by up to 6.1% compared to the next best semi-supervised method.
- 24 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 27 (15209210) , 1874-1886
An Active Multi-Target Domain Adaptation Strategy: Progressive Class Prototype Rectification
Compared to single-source to single-target (1S1T) domain adaptation, single-source to multi-target (1SmT) domain adaptation is more practical but also more challenging. In 1SmT scenarios, the significant differences in feature distributions between various target domains increase the difficulty for models to adapt to multiple domains. Moreover, 1SmT requires effective transfer to each target domain while maintaining performance in the source domain, demanding higher generalization capabilities from the model. In 1S1T scenarios, active domain adaptation methods improve generalization by incorporating a few target domain samples, but these methods are rarely applied in 1SmT due to potential sampling bias and outlier interference. To address this, we propose Progressive Prototype Refinement (PPR), an active multi-target domain adaptation method combining 1SmT with active learning to enhance cross-domain knowledge transfer. Specifically, an uncertainty assessment strategy is used to select representative samples from multiple target domains, forming a candidate set for model training. Based on the Lindeberg-Levy central limit theorem, we sample from a Gaussian distribution using corrected prototype statistics to augment the classifier's feature input, allowing the model to learn transitional information between domains. Finally, a mapping matrix is used for cross-domain alignment, addressing incomplete class coverage and outlier interference. Extensive experiments on multiple benchmark datasets demonstrate PPR's superior performance, with a 6.35% improvement on the PACS dataset and a 17.32% improvement on the Remote Sensing dataset.
- 24 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 27 (15209210) , 1874-1886
An Active Multi-Target Domain Adaptation Strategy: Progressive Class Prototype Rectification
Compared to single-source to single-target (1S1T) domain adaptation, single-source to multi-target (1SmT) domain adaptation is more practical but also more challenging. In 1SmT scenarios, the significant differences in feature distributions between various target domains increase the difficulty for models to adapt to multiple domains. Moreover, 1SmT requires effective transfer to each target domain while maintaining performance in the source domain, demanding higher generalization capabilities from the model. In 1S1T scenarios, active domain adaptation methods improve generalization by incorporating a few target domain samples, but these methods are rarely applied in 1SmT due to potential sampling bias and outlier interference. To address this, we propose Progressive Prototype Refinement (PPR), an active multi-target domain adaptation method combining 1SmT with active learning to enhance cross-domain knowledge transfer. Specifically, an uncertainty assessment strategy is used to select representative samples from multiple target domains, forming a candidate set for model training. Based on the Lindeberg-Levy central limit theorem, we sample from a Gaussian distribution using corrected prototype statistics to augment the classifier's feature input, allowing the model to learn transitional information between domains. Finally, a mapping matrix is used for cross-domain alignment, addressing incomplete class coverage and outlier interference. Extensive experiments on multiple benchmark datasets demonstrate PPR's superior performance, with a 6.35% improvement on the PACS dataset and a 17.32% improvement on the Remote Sensing dataset.