基于GMA光流网络的二维变形测量数字图像相关法初值估计

Initial Value Estimation of Digital Image Correlation Method for Two-Dimensional Deformation Measurement Based on GMA Optical Flow Network

提出一种利用深度学习提取散斑变形图像位移场的算法。对比几种主流光流网络模型在散斑图像上的计算精度,将全局运动聚合(GMA)光流网络引入数字图像相关法,使用光流网络模型预测的位移场作为亚像素迭代算法的初值,最后结合逆合成高斯-牛顿法计算散斑图像的亚像素位移场。结果表明,与几种主流的数字图像相关算法相比,GMA光流网络初值的计算精度和计算效率更高,适用于大变形测量,是一种兼顾计算速度和精度的变形测量算法。对木块材料进行压缩实验,结果充分显示了所提方法在散斑变形测量中的有效性。

Objective In the two-dimensional deformation measurement of speckle images,the initial value estimation of the digital image correlation method exerts great influence on the computational efficiency and accuracy of algorithms.The calculation accuracy and speed of sub-pixel displacement iterative search algorithms in digital image correlation methods depend on whether the initial value estimation provided by the integral pixel displacement calculation is reasonable or not,and its convergence radius is generally in the range of several pixels.Therefore,the initial value estimation provided in the integer pixel displacement search phase should be as close to the real value as possible to ensure that the iterative algorithm can converge quickly and accurately,otherwise,it may converge slowly or even fail in the iterative process.The traditional initial value estimation methods including the human-computer interaction method,Fourier transform method,and feature matching method,have some problems such as slow calculation speed and low calculation accuracy in the face of large deformation measurement and unclear speckle image features.Recently,the optical flow estimation network models in deep learning feature fast calculation speed,high calculation accuracy,and strong generalization in predicting motion displacement.We introduce the optical flow estimation network model in deep learning into the digital image correlation method and employ the displacement field predicted by the optical flow network as the initial value of the subpixel iterative algorithm.Finally,the inverse compositional Gauss-Newton method is adopted to calculate the displacement field of speckle deformation images.We hope that the strategy of introducing deep learning into the digital image correlation method can provide a new idea for speckle deformation measurement.Methods First of all,we compare the calculation accuracy of several optical flow network models of FlowNet2,PWCNet,RAFT,GMA,SeparableFlow,GMFlow,and FlowFormer,which have excellent performance on MPI Sintel test datasets on speckle images.Considering the calculation time,model size,and calculation accuracy,the GMA network model is chosen to provide initial value estimation for sub-pixel iterative algorithms.Then,a feature sampling module is added to the model for solving the problem that the GMA network needs to occupy a lot of GPU resources in highresolution speckle images,which can effectively reduce the occupation of GPU memory by adjusting the sampling step size.Additionally,the speckle images are utilized to generate many randomly deformed speckle datasets to retrain the model to enhance the generalization of the model in speckle deformation measurement.Finally,the GMA network is combined with the ICGN algorithm,and the performance of the algorithm is evaluated by simulated speckle deformation experiments and real wood block compression experiments.Results and Discussions After optimizing the sampling module,the computing resources needed in the model prediction continue to decrease with the increasing step size,and the sampling step can be reasonably selected by combining the hardware resources of the computer and calculation accuracy.After retraining in the speckle deformation dataset,the average endpoint error of the model in speckle images decreases by 14.76%.In large deformation measurement,the calculation accuracy of the proposed GMA-ICGN algorithm can still be kept at 0.01 pixel.Compared with the Fourier transform method and feature matching method in the initial value estimation algorithms,the computing speed of the GMA network has obvious advantages.In the wood block compression experiments,the GMA-ICGN algorithm successfully measures the displacement field and strain field of woodblock compression deformation.Conclusions The integral pixel displacement search algorithm in the digital image correlation method usually takes a long time.We propose a digital image correlation method based on GMA optical flow network.The reliable displacement initial value of speckle deformation images is obtained by the GMA network and then brought into ICGN iterative algorithm to accurately solve the displacement field,which can greatly improve the computational efficiency of the digital image phase method.At the same time,the GMA optical flow network is optimized and retrained,and the average endpoint error is reduced by 14.76%,which improves the prediction accuracy of the GMA network in the speckle images and reduces the GPU memory consumption by adjusting the step size of the network model.Through simulated speckle deformation experiments,a comparison between the calculation accuracy and efficiency of the proposed GMA-ICGN algorithm and the popular SIFT-ICGN algorithm,FFT-ICGN algorithm,Ncorr software,and DICe software proves that the proposed algorithm has higher computational efficiency.In addition,the proposed algorithm has similar accuracy to the SIFT-ICGN algorithm in large deformation scenes and can calculate the speckle deformation displacement field quickly and accurately.Furthermore,the proposed algorithm is applied to woodblock compression experiments,and the displacement field and strain field of woodblock compression deformation are measured successfully.