Spatial superresolution based on simultaneous dual PIV measurement with different magnification

A reconstruction framework based on proper orthogonal decomposition and the Bayesian estimation was designed for the spatial superresolution of a subsonic jet, and the simultaneous two PIV measurements of a subsonic jet with different magnifications were conducted for training and testing the framework. The measurement system successfully acquired paired particle images of broad and close-up views of the jet in the same plane, and low and high-resolution velocity fields were obtained. The artificial low-resolution velocity fields were also generated by average pooling of the measured high-resolution velocity fields, and the performance of the reconstruction framework was evaluated. The estimation accuracy of the proposed framework was compared with that of bicubic interpolation and machine learning-based reconstruction methods: convolutional neural network and downsampled skip-connection/multi-scale methods. The framework successfully reconstructed the high-resolution velocity field from the low-resolution velocity field of the artificial one and actually measured one. The minimum reconstruction error of the Bayesian estimation using actually measured low-resolution velocity field was 63%, outperforming bicubic interpolation. Although this reconstruction error of the proposed framework is almost the same as (slightly worse than) that of the neural network methods, its reconstruction process is clearer and simpler than the neural network method. The power spectra of turbulent kinetic energy showed that the proposed framework can accurately recover the original velocity field in a wide waveband compared to the other methods. Therefore, the proposed framework can be a superresolution method of experimental fluid dynamics.