Particle Image Velocimetry (PIV) for flow visualization has advanced with the integration of deep learning algorithms. These methods enable end-to-end processing, extracting dense flow fields directly from raw particle images. However, conventional deep learning-based PIV models, which predominantly rely on convolutional architectures, are limited in their ability to utilize contextual information and capture dependencies between pixels across sequential images, impacting the prediction accuracy. We introduce Twins-PIVNet, a deep learning framework for PIV optical flow estimation that leverages a spatial attention-based vision transformer architecture. Its self-attention mechanism captures multi-scale features of particle motion, significantly improving the dense flow field estimation. Trained on synthetic PIV datasets covering a wide range of flow conditions, Twins-PIVNet has been evaluated on both synthetic and experimental datasets, demonstrating superior accuracy and performance. In comparative studies, Twins-PIVNet outperforms existing optical flow and conventional methods, achieving accuracy improvements of 51% for backstep flow, 42% for DNS-turbulence, and 33% for surface quasi-geostrophic flow. Additionally, it also exhibits strong generalization on experimental PIV data, demonstrating robustness in handling real-world PIV uncertainties. Despite its attention mechanism, Twins-PIVNet maintains faster inference and training times compared to other PIV models, offering an optimal balance between complexity, efficiency, and performance.

A bi-directional digital holographic imaging system is presented that is designed to take images automatically out in the field. The main objective is to acquire sufficient information to be able to estimate the scattering phase function for different type of snowfall. The imaging system consists of a 3D-printed frame and two orthogonal telecentric imaging arms, one in the forward direction and one in the side scattering direction for which the reference arms are directed along different paths. All images are acquired using pulsed visible light.

We experimentally demonstrate the presence of a capillary bridge in the contact between an ice particle and a smooth aluminum surface at a relative humidity of approximately 50% and temperatures below the melting point. We conduct the experiments in a freezer with a controlled temperature and consider the mechanical instability of the bridge upon separation of the ice particle from the aluminum surface at a constant speed. We observe that a liquid bridge forms, and this formation becomes more pronounced as the temperature approaches the melting point. We also show that the separation distance is proportional to the cube root of the volume of the bridge. We hypothesize that the volume of the liquid bridge can be used to provide a rough estimate of the thickness of the liquid layer on the ice particle since in the absence of other driving mechanisms, some of the liquid on the surface must have been pulled to the bridge area. We show that the estimated value lies within the range previously reported in the literature. With these assumptions, the estimated thickness of the liquid layer decreases from nearly 56 nm at T = −1.7°C to 0.2 nm at T = −12.7°C. The dependence can be approximated with a power law, proportional to (TM − T)−β, where β < 2.6 and TM is the melting temperature. We further observe that for a rough surface, the capillary bridge formation in the considered experimental conditions vanishes.

Current optical flow-based neural networks for particle image velocimetry (PIV) are largely trained on synthetic datasets emulating real-world scenarios. While synthetic datasets provide greater control and variation than what can be achieved using experimental datasets for supervised learning, it requires a deeper understanding of how or what factors dictate the learning behaviors of deep neural networks for PIV. In this study, we investigate the performance of the recurrent all-pairs field transforms-PIV (RAFTs-PIV) network, the current state-of-the-art deep learning architecture for PIV, by testing it on unseen experimentally generated datasets. The results from RAFT-PIV are compared with a conventional cross-correlation-based method, Adaptive PIV. The experimental PIV datasets were generated for a typical scenario of flow past a circular cylinder in a rectangular channel. These test datasets encompassed variations in particle diameters, particle seeding densities, and flow speeds, all falling within the parameter range used for training RAFT-PIV. We also explore how different image pre-processing techniques can impact and potentially enhance the performance of RAFT-PIV on real-world datasets. Thorough testing with real-world experimental PIV datasets reveals the resilience of the optical flow-based method's variations to PIV hyperparameters, in contrast to the conventional PIV technique. The ensemble-averaged root mean squared errors between the RAFT-PIV and Adaptive PIV estimations generally range between 0.5–2 (px) and show a slight reduction as particle densities increase or Reynolds numbers decrease. Furthermore, findings indicate that employing image pre-processing techniques to enhance input particle image quality does not improve RAFT-PIV predictions; instead, it incurs higher computational costs and impacts estimations of small-scale structures.

Despite the significance of snow in various cryospheric, polar, and construction contexts, more comprehensive studies are required on its mechanical properties. In recent years, the utilization of μ CT has yielded valuable insights into snow analysis. Our objective is to establish a methodology for mapping density-dependent material properties for dry manufactured snow within the density range of 400–600 kg/m 3 utilizing μ CT imaging and step-wise, quasi-static, mechanical loading. We also aim to investigate the variations in the structural parameters of snow during loading. The three-dimensional (3D) structure of snow is captured using μ CT with 801 projections at the beginning of the experiments and at the end of each loading step. The sample is compressed at a temperature of − 18 o C using a constant rate of deformation (0.2 mm/min) in multiple steps. The relative density of the snow is determined at each load step using binary image segmentation. It varies from 0.44 in the beginning to nearly 0.65 at the end of the loading, which corresponds to a density range of 400–600 kg/m 3 . The estimated modulus and viscosity terms, obtained from the Burger’s model, show an increasing trend with density. The values of the Maxwell and Kelvin–Voigt moduli were found to range from 60 to 320 MPa and from 6 to 40 MPa, respectively. Meanwhile, the viscosity values for the Maxwell and Kelvin–Voigt models varied from 0.4 to 3.5 GPa-s, and 0.3–3.2 GPa-s, respectively, within the considered density range. In addition, Digital Volume Correlation (DVC) was used to calculate the full-field strain distribution in the specimen at each load step. The image analysis results show that, the particle size and specific surface area (SSA) do not change significantly within the studied range of loading and densities, while the sphericity of the particles is increased. The grain diameter ranges from approximately 100 μ m to nearly 400 μ m, with a mode of nearly 200 μ m. The methodology presented in this study opens up a path for an extensive statistical analysis of the material properties by experimenting more snow samples.

The mechanical properties of natural snow play a crucial role in understanding glaciers, avalanches, polar regions, and snow-related constructions. Research has concentrated on how the mechanical properties of snow vary, primarily with its density; the integration of cutting-edge techniques like micro-tomography with traditional loading methods can enhance our comprehension of these properties in natural snow. This study employs CT imaging and uniaxial compression tests, along with the Digital Volume Correlation (DVC) to investigate the density-dependent material properties of natural snow. The data from two snow samples, one initially non-compressed (test 1) and the other initially compressed (test 2), were fed into Burger’s viscoelastic model to estimate the material properties. CT imaging with 801 projections captures the three-dimensional structure of the snow initially and after each loading step at -18C, using a constant deformation rate (0.2 mm/min). The relative density of the snow, ranging from 0.175 to 0.39 (equivalent to 160–360 kg/m), is determined at each load step through binary image segmentation. Modulus and viscosity terms, estimated from Burger’s model, exhibit a density-dependent increase. Maxwell and Kelvin–Voigt moduli range from 0.5 to 14 MPa and 0.1 to 0.8 MPa, respectively. Viscosity values for the Maxwell and Kelvin–Voigt models vary from 0.2 to 2.9 GPa-s and 0.2 to 2.3 GPa-s within the considered density range, showing an exponent between 3 and 4 when represented as power functions. Initial grain characteristics for tests 1 and 2, obtained through image segmentation, reveal an average Specific Surface Area (SSA) of around 55 1/mm and 40 1/mm, respectively. The full-field strain distribution in the specimen at each load step is calculated using the DVC, highlighting strong strain localization indicative of non-homogeneous behavior in natural snow. These findings not only contribute to our understanding of natural snow mechanics but also hold implications for applications in fields such as glacier dynamics and avalanche prediction.

This study focuses on utilizing image techniques for river velocity measurement, with a specific emphasis onnatural surface floating patterns. Employing a multi-camera system, we conducted 3D measurements on riversurfaces, including surface velocity and water surface reconstruction. A pattern-based tracking approach hasbeen adopted to improve the performance of image measurements on different types of natural floating tracers.The study employs the following approaches: 3D Lagrangian Pattern Tracking Velocimetry (3D-LPTV), 2DLagrangian Pattern Velocimetry (2D- LPTV), and Large-scale Particle Image Velocimetry (LSPIV), for surfacevelocity estimation. The outcomes revealed that all three approaches yielded consistent results in terms ofaveraged velocity. However, the LSPIV method produced about two times higher uncertainty in measured velocitiescompared to the other methods. A strategy to assess the quality of river surface patterns in velocityestimation is presented. Specifically, the sum of squared interrogation area intensity gradient (SSIAIG) was foundto be strongly correlated with measurement uncertainty. Additionally, a term related to the peak sidelobe ratio(PSR) of the cross-correlation map was found as an effective constraint, ensuring the image-tracking processachieves high reliability. The precision of measurements increases corresponding to the increase of image intensitygradient and PSR.

We propose to consider the speckle motions from induced phase gradients to provide a criterion for accurate image refocusing in digital holographic interferometry. Experiments confirm the theory.