流场测速中基于深度卷积神经网络的光学畸变校正技术

基于光学成像的流场测量技术，如粒子图像测速技术（PIV），易受到因流体中折射率的不均匀性或晃动的介质边界引起的光学畸变而带来的影响。这些畸变会使得示踪粒子在图像上的位置分布产生误差且严重影响图像清晰度，从而增大流场速度测量的误差。为了提高光学流场速度测量的测量精度，自适应光学系统可以应用于其中去校正光学畸变。基于图像流场测量中的光学像差具有频率高，动态范围大，空间分辨率高等特点，对于这一应用场景，基于波前校正器件的自适应光学系统受到了器件本身性能的影响。基于深度学习的自适应光学技术在流场测量中的应用，建立了一种基于深度神经网络的无波前校正器件自适应光学校正技术，以深度神经网络代替传统的波前校正器件，用于粒子图像测速技术中的光学畸变校正。为了生成神经网络所需要的训练和测试数据集，搭建了可以实现波前测量的粒子图像测速实验平台，分析并建立了光学畸变在粒子图像上的图像退化模型。最后，以校正后PIV图像的校正效果和流场速度测量结果作为评价标准，对所建立神经网络的畸变校正性能进行了分析。

Aberration correction for flow velocity measurements using deep convolutional neural networks

Optical imaging-based flow measurement techniques, like particle image velocimetry, are vulnerable to optical distortions caused by inhomogeneous refractive index or fluctuating phase boundaries. These distortions can lead to blurred particle images and uncertain tracer particle position assignment, resulting in a degradation of velocity measurement accuracy. In order to improve the measurement accuracy, adaptive optics system can be applied to correct distortions. For imaging metrology in fluid mechanics, the optical distortions have features of large frequency range, high spatial frequency and large dynamic range. Actuator-based approaches are limited by its performances. In our work, a novel intelligent adaptive optic system was applied to flow measurement, a learning-based aberration correction method without wavefront corrector was demonstrated, which was used to correct distortions in imaging-based flow measurement. A particle image velocimetry setup which can measure wavefront aberration was built to generate training and test dataset for deep neural network, and also the distortion caused PIV image degradation model. The correction performance of the trained neural network was quantitatively evaluated by corrected PIV image quality and flow measurement result.