Convolutional Neural Network Based DIC Method for Natural Texture

Digital image correlation (DIC) is a non-contact full-field displacement field measurement method based on image analysis. Existing DIC methods rely on manually produced highprecision and high-quality speckle images, which are not only time-consuming and labour-intensive but also demanding, and easy to fall off in the actual measurement engineering. Therefore, in this paper, a convolutional neural network is proposed to measure the natural texture displacement field on the surface of metallic materials. The experiments and results are analysed by comparing with existing deep learning methods on a self-built dataset. The results show that the method in this study has the best measurement accuracy and the fastest measurement speed. In addition, the feasibility of this research method in practical applications is verified by real stretching experiments.