Shengyu Tang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shengyu Tang has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- Changdong Yu
College of Artificial Intelligence, Dalian Maritime University, Dalian 116026, China
3 shared publications - Shuaiyu Bao
College of Shipping Economics and Management, Dalian Maritime University 5 , Dalian 116026
2 shared publications - Yang Han
College of Shipbuilding Engineering, Harbin Engineering University 6 , Harbin 150001
2 shared publications - Yaohui Lyu
Faculty of Information Science and Engineering, Ocean University of China, Qingdao, China
2 shared publications - Fei Zhou
College of Oceanography and Space Informatics, China University of Petroleum (East China), Qingdao, China
1 shared publication - Huijie Ma
College of Engineering, Ocean University of China, Qingdao, China
1 shared publication
Latest Publications
- Open Access
- 26 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. PP (19391404) , 1-20
A Dual-Branch Framework for Robust Target Tracking in Underwater Acoustic Scenarios
Robust underwater acoustic target tracking is critical for a wide range of marine applications, including resource exploration, subsea infrastructure inspection, and underwater security. However, complex underwater environments characterized by severe acoustic noise, highly non-stationary target signatures, and the self-motion of sensing platforms pose major challenges to existing tracking-by-detection pipelines. To address these issues, we propose UADSORT, a novel and efficient framework tailored for underwater acoustic tracking. At its core, UADSORT adopts a Dual-Stream Decoupled Perception Architecture (DDPA), which enhances perception in parallel while explicitly separating platform-induced motion from target dynamics. Specifically, we introduce a Sonar Echo Star-Enhanced Detector (SES-Det) that leverages a star -shaped feature interaction mechanism to achieve high-accuracy detection in low-SNR sonar imagery. Meanwhile, we further introduce an Acoustic Robust Appearance Extractor (ARAE) to learn discriminative identity features that remain stable under pose variations and partial occlusions. In addition, the framework integrates an explicit motion compensation module to mitigate trajectory distortions caused by platform perturbations. Finally, an Adaptive Clue Fusion Framework (ACFF) dynamically combines motion and appearance cues to enable robust data association. Extensive evaluations on the RGBS50-Sonar benchmark and in Qingdao sea-trial experiments demonstrate that UADSORT achieves state-of-the-art performance.
- 26 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. PP (19391404) , 1-20
Open AccessA Dual-Branch Framework for Robust Target Tracking in Underwater Acoustic Scenarios
Robust underwater acoustic target tracking is critical for a wide range of marine applications, including resource exploration, subsea infrastructure inspection, and underwater security. However, complex underwater environments characterized by severe acoustic noise, highly non-stationary target signatures, and the self-motion of sensing platforms pose major challenges to existing tracking-by-detection pipelines. To address these issues, we propose UADSORT, a novel and efficient framework tailored for underwater acoustic tracking. At its core, UADSORT adopts a Dual-Stream Decoupled Perception Architecture (DDPA), which enhances perception in parallel while explicitly separating platform-induced motion from target dynamics. Specifically, we introduce a Sonar Echo Star-Enhanced Detector (SES-Det) that leverages a star -shaped feature interaction mechanism to achieve high-accuracy detection in low-SNR sonar imagery. Meanwhile, we further introduce an Acoustic Robust Appearance Extractor (ARAE) to learn discriminative identity features that remain stable under pose variations and partial occlusions. In addition, the framework integrates an explicit motion compensation module to mitigate trajectory distortions caused by platform perturbations. Finally, an Adaptive Clue Fusion Framework (ACFF) dynamically combines motion and appearance cues to enable robust data association. Extensive evaluations on the RGBS50-Sonar benchmark and in Qingdao sea-trial experiments demonstrate that UADSORT achieves state-of-the-art performance.
- 1 April 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (4)
Probabilistic physics-constrained reconstruction of three-dimensional flow fields
Accurate reconstruction of three-dimensional flow fields is essential for resolving fine-scale flow structures and understanding unsteady flow physics in ship and ocean engineering applications. However, while direct numerical simulation provides high-fidelity solutions of the Navier–Stokes equations, its prohibitive computational cost limits its applicability to large-scale and high-Reynolds-number flows. Reduced-order and data-driven reconstruction approaches offer improved efficiency, yet often fail to preserve inter-slice coherence, fine-scale vortical structures, and physical consistency when extrapolating beyond the training regime. To address these challenges from a physical-consistency perspective, this study investigates a physics-constrained framework for three-dimensional flow-field reconstruction. The governing physical constraints are embedded into the reconstruction process through a probabilistic formulation derived from the vorticity–velocity representation of the Navier–Stokes equations, enabling improved enforcement of incompressibility and dynamical consistency. A spatiotemporal reconstruction architecture is constructed to capture multi-scale spatial features and temporal evolution characteristics of unsteady flows, while a physics-informed attention mechanism is employed to enhance the recovery of fine-scale flow structures in regions with strong spatial gradients. The proposed framework is assessed using turbulent flow data from the Johns Hopkins Turbulence Database, transient boundary layer flows, and experimental tomographic particle image velocimetry measurements. Quantitative comparisons with high-fidelity reference data demonstrate that the physics-constrained reconstruction significantly improves the preservation of fine-scale flow structures and extrapolation performance compared with conventional deterministic approaches, indicating its effectiveness for physically consistent three-dimensional flow-field reconstruction at reduced computational cost.
- 1 April 2026
- Published by AIP Publishing in Physics of Fluids
- Vol. 38 (4)
Probabilistic physics-constrained reconstruction of three-dimensional flow fields
Accurate reconstruction of three-dimensional flow fields is essential for resolving fine-scale flow structures and understanding unsteady flow physics in ship and ocean engineering applications. However, while direct numerical simulation provides high-fidelity solutions of the Navier–Stokes equations, its prohibitive computational cost limits its applicability to large-scale and high-Reynolds-number flows. Reduced-order and data-driven reconstruction approaches offer improved efficiency, yet often fail to preserve inter-slice coherence, fine-scale vortical structures, and physical consistency when extrapolating beyond the training regime. To address these challenges from a physical-consistency perspective, this study investigates a physics-constrained framework for three-dimensional flow-field reconstruction. The governing physical constraints are embedded into the reconstruction process through a probabilistic formulation derived from the vorticity–velocity representation of the Navier–Stokes equations, enabling improved enforcement of incompressibility and dynamical consistency. A spatiotemporal reconstruction architecture is constructed to capture multi-scale spatial features and temporal evolution characteristics of unsteady flows, while a physics-informed attention mechanism is employed to enhance the recovery of fine-scale flow structures in regions with strong spatial gradients. The proposed framework is assessed using turbulent flow data from the Johns Hopkins Turbulence Database, transient boundary layer flows, and experimental tomographic particle image velocimetry measurements. Quantitative comparisons with high-fidelity reference data demonstrate that the physics-constrained reconstruction significantly improves the preservation of fine-scale flow structures and extrapolation performance compared with conventional deterministic approaches, indicating its effectiveness for physically consistent three-dimensional flow-field reconstruction at reduced computational cost.
- 1 October 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (10)
An efficient deep optical flow learning approach for two-dimensional particle image velocimetry
In the particle image velocimetry (PIV) community, velocity fields derived from high-speed camera sensors are invaluable for gaining deeper insight into fluid flow dynamics. However, although current deep learning approaches achieve high accuracy, the most effective models often incur high computational costs, limiting their applicability to PIV estimation. To address this limitation, we propose the recurrent feature pyramid aggregation PIV method (RFPA-PIV), a novel deep learning framework for optical flow estimation designed for two-dimensional (2D) flow fields. RFPA-PIV integrates efficient InceptionNeXt convolutional blocks within a pyramid-based recurrent architecture, reducing computational complexity while maintaining accuracy. Multi-scale features from the pyramid are aggregated and fed into a four-dimensional correlation volume to compute visual similarity. Furthermore, RFPA-PIV was trained on a diverse 2D PIV dataset covering a wide range of flow conditions and evaluated on both synthetic benchmarks and real zebrafish PIV images. Comparative experiments show that RFPA-PIV outperforms existing conventional and deep learning-based methods, achieving accuracy improvements of 80% in cylinder flow, 69% in backstep flow, and 63% in surface quasi-geostrophic flow. It also exhibits strong generalization on real zebrafish PIV data, demonstrating robustness in handling real-world 2D measurement. Compared to other PIV approaches, RFPA-PIV significantly reduces graphics processing unit memory consumption and enables faster inference, effectively balancing model complexity, computational efficiency, and estimation accuracy.
- 1 October 2025
- Published by AIP Publishing in Physics of Fluids
- Vol. 37 (10)
An efficient deep optical flow learning approach for two-dimensional particle image velocimetry
In the particle image velocimetry (PIV) community, velocity fields derived from high-speed camera sensors are invaluable for gaining deeper insight into fluid flow dynamics. However, although current deep learning approaches achieve high accuracy, the most effective models often incur high computational costs, limiting their applicability to PIV estimation. To address this limitation, we propose the recurrent feature pyramid aggregation PIV method (RFPA-PIV), a novel deep learning framework for optical flow estimation designed for two-dimensional (2D) flow fields. RFPA-PIV integrates efficient InceptionNeXt convolutional blocks within a pyramid-based recurrent architecture, reducing computational complexity while maintaining accuracy. Multi-scale features from the pyramid are aggregated and fed into a four-dimensional correlation volume to compute visual similarity. Furthermore, RFPA-PIV was trained on a diverse 2D PIV dataset covering a wide range of flow conditions and evaluated on both synthetic benchmarks and real zebrafish PIV images. Comparative experiments show that RFPA-PIV outperforms existing conventional and deep learning-based methods, achieving accuracy improvements of 80% in cylinder flow, 69% in backstep flow, and 63% in surface quasi-geostrophic flow. It also exhibits strong generalization on real zebrafish PIV data, demonstrating robustness in handling real-world 2D measurement. Compared to other PIV approaches, RFPA-PIV significantly reduces graphics processing unit memory consumption and enables faster inference, effectively balancing model complexity, computational efficiency, and estimation accuracy.
- 5 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 73 (6) , 8576-8583
Reparameterization Causal Convolutional Network for Automatic Modulation Classification
With the proliferation of wireless technologies in vehicular networks, robust automatic modulation classification (AMC) has become crucial for optimizing spectrum utilization and maintaining reliability. However, AMC in dynamic vehicular channels poses significant challenges for traditional machine learning techniques. This paper proposes a novel CNN-based approach named Reparameterization Causal Convolutional Network (RepCCNet) to achieve highly accurate and noise-robust AMC performance. RepCCNet incorporates causal convolutions and structural reparameterization techniques to extract long-term time-domain features. A bottleneck structure with channel attention dynamically calibrates feature channels, retaining only helpful information. Multi-sample dropout is integrated during training to improve generalization capability. We demonstrate RepCCNet's state-of-the-art classification accuracy on the two widely used datasets, RadioML 2016.10a and RadioML 2018.01a, across varying signal-to-noise ratios. Compared to existing methods, RepCCNet achieves highly competitive results compared to state-of-the-art approaches, utilizing fewer than 40k parameters. Ablation studies validate the contributions of the proposed architectural innovations. This work represents a significant advancement toward developing deep learning solutions for robust wireless signal classification tasks.
- 5 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 73 (6) , 8576-8583
Reparameterization Causal Convolutional Network for Automatic Modulation Classification
With the proliferation of wireless technologies in vehicular networks, robust automatic modulation classification (AMC) has become crucial for optimizing spectrum utilization and maintaining reliability. However, AMC in dynamic vehicular channels poses significant challenges for traditional machine learning techniques. This paper proposes a novel CNN-based approach named Reparameterization Causal Convolutional Network (RepCCNet) to achieve highly accurate and noise-robust AMC performance. RepCCNet incorporates causal convolutions and structural reparameterization techniques to extract long-term time-domain features. A bottleneck structure with channel attention dynamically calibrates feature channels, retaining only helpful information. Multi-sample dropout is integrated during training to improve generalization capability. We demonstrate RepCCNet's state-of-the-art classification accuracy on the two widely used datasets, RadioML 2016.10a and RadioML 2018.01a, across varying signal-to-noise ratios. Compared to existing methods, RepCCNet achieves highly competitive results compared to state-of-the-art approaches, utilizing fewer than 40k parameters. Ablation studies validate the contributions of the proposed architectural innovations. This work represents a significant advancement toward developing deep learning solutions for robust wireless signal classification tasks.