Junpeng Zhu

Junpeng Zhu

Auto-generated profile, based on publicly available information

Junpeng Zhu


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 October 2025 in Physics of Fluids
Verified
Journal Article
Closed access
STShengyu Tang
Shengyu Tang
SBShuaiyu Bao
Shuaiyu Bao
CYChangdong Yu
Changdong Yu
YHYang Han
Yang Han
JZJunpeng Zhu
Junpeng Zhu
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 July 2025 in Physics of Fluids
Verified
Journal Article
Closed access
YCYiming Cao
Yiming Cao
CYChangdong Yu
Changdong Yu
PLPan Li
Pan Li
CRChenyi Rong
Chenyi Rong
JZJunpeng Zhu
Junpeng Zhu
SBShuaiyu Bao
Shuaiyu Bao
Particle image velocimetry (PIV) estimates velocity fields by capturing particle movements in consecutive images, holding significant importance in marine and naval fields. Thus, acquiring high-quality multi-frame particle images via PIV estimation is crucial. However, current PIV image preprocessing methods only focus on enhancing two-frame particle image pairs, ignoring the spatiotemporal correlation in continuous multi-frame particle images. Considering complex noise during continuous particle image capture and spatiotemporal correlations between image pairs, we designed the spatiotemporal feature interaction network (SFINet) for particle image preprocessing. First, in terms of spatial features, given particles' tiny, non-uniform, and highly similar features, we use an encoder–decoder with the mixed feature enhancement module and the adaptive fine-grained channel attention module to extract, enhance, and reconstruct particle detail features. Second, in terms of temporal features, three-dimensional convolution layers leverage spatiotemporal correlations in continuous particle images for feature fusion, completing image enhancement. For real-world noise, we autonomously generated a challenging image enhancement dataset, including low contrast, Gaussian noise (various concentrations), light intensity noise, and real underwater interference backgrounds to simulate real underwater environments. The results show that the proposed SFINet can effectively address the noise interference in both synthetic and real images. On the test set, it achieves particle image enhancement with the average peak signal-to-noise ratio reaching 36.255 and structural similarity index measure hitting 0.9843 between the denoised and ground-truth images, demonstrating its remarkable performance in handling noise and improving image quality.

1 August 2025 in Physics of Fluids
Verified
Journal Article
Closed access
CGChunyu Guo
Chunyu Guo
YWYonghao Wang
Yonghao Wang
YHYang Han
Yang Han
YFYiwei Fan
Yiwei Fan
YWYanyuan Wu
Yanyuan Wu
JZJunpeng Zhu
Junpeng Zhu
The pump-jet propulsor (PJP), as a new type of propulsion system for underwater vehicles characterized by high efficiency, low noise, and high critical speed, plays a significant role in advancing marine engineering. Obtaining complete internal flow field data with high spatial and temporal resolution is crucial for revealing the mechanisms underlying mechanical-acoustic coupling evolution. During experiments, when using particle image velocimetry (PIV) to measure the internal flow fields of pump jets, factors such as complex structural occlusions, wall reflections, and uneven particle distribution can lead to random missing flow field information. Therefore, we developed a U-shaped attention-driven deformable generative adversarial network (UAD-GAN) framework based on data-driven techniques and deep learning. First, we obtained datasets of internal pump-jet flow fields at various advanced coefficients using computational fluid dynamics (CFD), which serves as the sample information for network training. Next, our framework employs an attention transfer network to establish a multi-scale feature fusion mechanism that facilitates dynamic interaction between shallow and deep flow features. By integrating deformable convolutions, our approach captures the complex and irregular boundary flow characteristics of pump-jet components, achieving high-precision end-to-end reconstruction of missing flow fields. To evaluate the robustness and generalization capability of our model, we systematically assess its performance in compensating for missing flow fields under four different missing modes. Additionally, we conduct a comprehensive comparison of UAD-GAN, deep convolutional neural networks (DCNNs), and autoencoder-based CNNs (CNN-AEs) from multiple perspectives, including compensation error, single-frame inference speed, and model training duration. The final results demonstrate that the UAD-GAN framework maintains a root mean square error below 5% when compensating for missing flow fields in pump jets, effectively achieving compensation for complex structures.

1 September 2025 in Physics of Fluids
Verified
Journal Article
Closed access
CWChunhui Wang
Chunhui Wang
JZJunpeng Zhu
Junpeng Zhu
YHYang Han
Yang Han
YFYiwei Fan
Yiwei Fan
CYChangdong Yu
Changdong Yu
YWYonghao Wang
Yonghao Wang
In experimental measurements of flow fields, the acquired data often fail to adequately resolve the fine structural characteristics of complex flows due to limitations in current technical conditions and measurement techniques. This constraint is particularly pronounced in analyzing turbulent random fluctuations and identifying small-scale vortices, hindering in-depth investigations into flow field dynamics. To address this challenge, this study proposes a novel super-resolution reconstruction method for flow fields based on pooling decomposition and a vortex attention mechanism. By introducing an innovative attention mechanism for high-order flow field quantities, the model effectively focuses on vortices within the flow field and extracts high-order information to guide velocity field reconstruction. Simultaneously, a pooling decomposition strategy is employed to specifically learn the null-space components of the flow field while preserving range-space components to enhance reconstruction performance. Notably, the introduction of a physics-constrained loss function ensures kinetic energy spectrum consistency in the reconstructed results. Experimental results demonstrate that the proposed model significantly improves the accuracy of flow field data reconstruction, exhibiting outstanding performance in turbulent kinetic energy spectrum recovery and small-scale vortex reconstruction, along with excellent cross-condition generalization capability. This research provides a new technical approach to overcome the resolution limitations of experimental measurements and achieve high-precision analysis of complex flow structures.

1 August 2025 in Physics of Fluids
Verified
Journal Article
Closed access
CWChunhui Wang
Chunhui Wang
JZJunpeng Zhu
Junpeng Zhu
YHYang Han
Yang Han
YFYiwei Fan
Yiwei Fan
CYChangdong Yu
Changdong Yu
YWYonghao Wang
Yonghao Wang
Three-dimensional particle field reconstruction serves as the foundation for three-dimensional velocity measurements in turbulent flows, and its accuracy directly determines the measurement reliability of particle image velocimetry systems. To address the typical limitations of conventional algebraic reconstruction techniques—such as computational inefficiency, poor adaptability to high-concentration fields, limited noise resistance, and particle morphology distortion—this study proposes the latent space back-projection particle reconstruction network. The framework establishes an implicit spatial mapping mechanism, employing differentiable projection operators to construct a nonlinear relationship between two-dimensional multi-view image feature tensors and three-dimensional implicit tensor space, thereby achieving end-to-end voxel feature reconstruction. Experimental validation demonstrates that the proposed algorithm significantly outperforms traditional algebraic reconstruction techniques in reconstruction fidelity on standard test datasets, with its depth-wise intensity distribution exhibiting superior Gaussian statistical characteristics. By leveraging shared encoding and an implicit space projection mechanism, the framework achieves single-pass reconstruction of 2083 voxels, substantially reducing graphics processing unit memory usage. This provides an efficient and reliable voxel reconstruction solution for three-dimensional measurements in complex flow fields.