Neutralizing the impact of heat haze on digital image correlation measurements via deep learning

Abstract

Digital image correlation (DIC) techniques have shown excellent capabilities in the deformation measurements of materials and structures at high temperatures. However, as a crucial challenge for the high-temperature DIC measurements, heat haze, caused by uneven distributions of temperatures and refractive indices, can lead to considerable distortions to the captured images and severely decrease the measurement accuracy. Inspired by a recently established atmospheric turbulence neutralization neural network called TSR-WGAN, we propose a deep learning-based approach to neutralize the effect of heat haze on high-temperature DIC measurements. Specifically, the original distorted speckle images obtained in the experiment are fed into the TSR-WGAN network twice to obtain distortion-corrected speckle images. By processing these corrected speckle images with a conventional DIC algorithm, displacement and strain fields with mitigated heat haze effects can be determined. Both simulation and real experiments have been conducted to assess the proposed method. Results clearly show the effectiveness and robustness of the proposed method in correcting heat haze effects. One of the experiments particularly shows that the displacement fluctuations caused by the heat haze can be reduced by approximately an order of magnitude in the u direction and by a factor of five in the v direction. The proposed technique provides a much-needed tool for neutralizing the impact of heat haze on the DIC measurements under a high-temperature environment.