Work

Efficient and robust deformation measurement based on unsupervised learning

Year: 2024

Type: article

Abstract: Digital image correlation (DIC) as a non-contact deformation measurement technique is widely used in various engineering fields. Recently, satisfactory results have been achieved by deep learning-based DIC methods, which rely on synthetic datasets for supervised training. However, generalization to real scenes remains challenging due to the inherent differences between synthetic and real images and the limited variability of synthetic datasets. For this problem, this paper proposes the unsupervised learning based DIC method. We designed a pyramid structure-based network for deformation measurement called UnDIC-Net, and proposed a Patch-based Zero-Normalized Sum of Squared Differences (Patch-ZNSSD) as similarity measure to construct unsupervised loss. Meanwhile, we collected speckle image from real scenes and built an unlabeled dataset to train UnDIC-Net. Adequate experiments show that UnDIC-Net performs well in small deformation measurements, and for larger deformation measurements (>100 pixels), some traditional methods fail while UnDIC-Net still performs well.

Source: Measurement

DOI: 10.1016/j.measurement.2024.115908

Authors: Jiashuai Yang, Yahong Feng, Lianpo Wang

Institution: Northwestern Polytechnical University

Language: English

Topic: Optical measurement and interference techniques

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $3,650

Funders: National Natural Science Foundation of China, Natural Science Basic Research Program of Shaanxi Province