Complex-background-oriented schlieren technology based on convolutional neural networks.

Area of Science:

• Fluid dynamics
• Optical measurement techniques

Background:

• Background-oriented schlieren (BOS) is crucial for visualizing transparent flows.
• Traditional algorithms face challenges with complex, noisy backgrounds in large-scale flow fields.

Purpose of the Study:

• To develop a robust model for accurate flow field measurement and visualization in challenging environments.
• To improve the performance of BOS techniques in the presence of complex and noisy backgrounds.

Main Methods:

• Physics-based noise augmentation applied to the IRR-PWCNet model, creating the robust-IRR model.
• Utilizing a bi-conjugate gradient (BiCG) method with ASDI-based initial values for flow field and wavefront reconstruction.

Main Results:

• The robust-IRR model reduced endpoint error (EPE) by 43.06% in noisy remote sensing backgrounds.
• Error reduction of 11.72% to 14.20% was observed at different PSNR levels compared to IRR without noise enhancement.
• A 7.47% error reduction was achieved in flow field and wavefront reconstruction using the combined BiCG and ASDI method.

Conclusions:

• The robust-IRR model provides a highly efficient and accurate solution for flow field measurement and visualization.
• The proposed method demonstrates significant improvements in robustness and precision, even in complex and noisy environments.
• This approach enables high-precision wavefront reconstruction, expanding the applicability of BOS techniques.