Fig. 5: The performance evaluation of RTU-Net on LF-PIV imaging of square lid-driven cavity flow.
a The diagram illustrates an experiment with five particles placed vertically, spaced apart by 5 mm intervals. b The light-field image captured in (a) is reconstructed by refocusing (green) and RTU-Net (red), respectively. The central position of each particle is extracted, in which blue is the ideal position. c In Cartesian coordinates, the distance between the reconstructed particle position xi and the ideal position \({{\boldsymbol{x}}^{i}_{ideal}}\) in space is expressed as a tensor \({\sqrt{{\sum}^{3}_{i=1}\left({{\boldsymbol{x}}_{i}-{\boldsymbol{x}}^{i}_{ideal}}\right)^2}}\). d The histogram shows the absolute deviation between the particle’s actual and ideal location in the z direction from the perspective of x-z. e Schematic drawing of setups for square lid-driven cavity flow measurement (upper) and the secondary vortex in the imaging area (bottom). f The flow velocity vector fields acquired by refocusing, RTU-Net, and Fluent. The central point of the captured secondary vortex is marked by a purple circle, while the flow separation region is indicated by a blue box. g The streamline visualization derived from refocusing, RTU-Net, and Fluent. h, The velocity error provided by refocusing and RTU-Net when compared with Fluent-simulated velocity vectors, n = 5000
