Lightweight dense U-Net for 3D tracer particle field reconstruction in synthetic aperture PIV

Synthetic aperture particle image velocimetry (SAPIV) has emerged as an advanced technique for investigating complex fluid dynamics, providing nonintrusive flow measurements with an extended depth of field, robustness at high particle concentrations, and improved three-dimensional resolutions. In SAPIV, accurate reconstruction of three-dimensional tracer particle fields is critical for ensuring the reliability and precision of flow measurements. In this study, a three-dimensional dense U-Net network is proposed to extract tracer particles from the synthetic aperture refocused image volume. Designed with a lightweight architecture, the network eliminates down-sampling and promotes feature reuse via dense blocks and skip connections. The 1×1×1 structure of the dense blocks allows for the reduced computational load while maintaining high spatial resolution. Validation in simulated vortex ring and uniform flow fields confirmed the method's accuracy in reconstructing tracer particle and displacement vector fields. Quantitative assessments using reconstruction quality Q and 3D structural similarity demonstrated that the method achieves high-quality reconstruction of 3D tracer particle fields. Furthermore, a real experiment involving flow around a circular cylinder was conducted to demonstrate the practical applicability of the method.