Combining deep learning and soft interferometric nanostrain sensor to probe nanoscale spatiotemporal surface deformation

Investigating the spatiotemporal deformation of soft substrate induced by cavitation and shock waves is necessary to understand the underlying mechanisms and visualize surface damage. Our group has recently proposed the use of a soft interferometric nanostrain sensor to measure nanoscale surface deformation of soft PDMS induced by stress acting on the surface. However, this methodology is plagued by the time-consuming denoising of the raw interferograms. Herein, we propose the use of a deep learning model to simplify the processing pipeline and achieve a significantly faster data analysis. A transferable hybrid network comprising both ResNet and ResNeXt modules embedded in a U-Net architecture is trained to achieve a 4000-fold speedup in denoising raw interferograms. Furthermore, we introduced the deep learning-enhanced cross-correlation filter and iterative learning to obtain denoised interferograms that are superior to the ground truth used to train the deep learning model, thereby enabling the correction of defects present in the ground truth denoised interferograms. The applicability of this methodology is demonstrated by investigating the PDMS surface deformation caused by an evaporating sessile DI water drop and a free-fall drop impingement onto the surface.