Convolutional neural network-based image denoising for synchronous measurement of temperature and deformation at elevated temperature

Abstract

Non-contact measurement method at elevated temperatures has been widely studied, which provides an efficient means for evaluating the properties of high-temperature materials. However, such high-temperature environment-induced challenges as strong light radiation and air disturbance may cause a huge quality decrease to the captured images based on the optical systems. As an important application of neural networks, Denoising Convolutional Neural Networks (DNCNN) has shown a considerable performance in conventional image denoising. In this work, an optimized DNCNN method based on the sub-region processing and transfer learning was proposed and applied in the high-temperature measurement. The thermal heating experiment of the SiC material was carried out to validate the applicability and accuracy of the method, while the synchronous measurement of temperature and deformation is realized. Simulation results based on the finite element method (FEM) agree well with the calculated data, indicating the proposed model can improve the measurement accuracy.