基于SwinTransformer的位移与应变测量

打开文本图片集

关键词：数字图像相关;SwinTransformer;高频位移；多尺度特征提取;应变测量中图分类号：TP394.1;TH691.9 文献标识码：Adoi：10.37188/OPE.20253315.2455 CSTR：32169.14.OPE.20253315.2455

Displacement strain measurement based on Swin Transformer

DAN，YANG，LI

（ 23O6ol，China） * Corresponding author， E-mail： guihualil@l63. com

Abstract： To address the limited high-frequency displacement capture capability poor adaptability traditional digital image correlation （DIC） methods that rely on fixed subsets，a strain- displacement synchronous prediction network based on the Swin Transformer is proposed. The model comprises two subnetworks，ST-DIC-d （displacement prediction） ST-DIC-s （strain prediction）. A hierarchical encoder-decoder Swin Transformer is employed for multi-scale feature extraction，enabling effective capture local global image features via attention mechanisms skip connections to recover the displacement field.A trainable strain computation layer then derives strain components directly from the predicted displacement field. Evaluation on the DIC Challnge dataset shows that ST-DIC reduces the average absolute error in vertical displacement prediction by 21% relative to traditional DIC CNN-based approaches.In tensile experiments，ST-DIC attains an 8% improvement in maximum strain accuracy within highgradient regions yields more finely resolved measurements. The proposed method demonstrates enhanced reliability accuracy for measuring complex deformations.

Key words： digital image corelation；swin transformer; high-frequency displacement；multi-scale featureextraction；strainmeasurement

1引言

数字图像相关（DigitalImageCorrelation，DIC）是一种在20世纪80年代发展起来的非接触式全场光学测量技术，可精确测量目标表面的位移和应变。（剩余16837字）