SEM-BSE micrographs of UO2 for grain boundary segmentation (model parameters and datasets)

This repository contains the data, trained deep learning model parameters and SEraMic grain boundary detection results presented in the manuscript entitled 'A deep-learning approach to grain boundary detection in backscattered electron images'. The folder Data contains 12 experimental datasets, (Di)i=1,...,6 and (Ti)i=1,...,6, of SEM-BSE images of uranium dioxide samples and their corresponding grain boundary images and a synthetic dataset. (Di)i=1,...,6 and (Ti)i=1,...,5 were obtained from [1, 2] and T6 was provided from [3]. Each Di dataset contains one SEM-BSE image and the corresponding grain boundary image, and each SEM-BSE image corresponds to an independent region of interest (ROI). Each Ti dataset contains 10-12 SEM-BSE images of the same ROI acquired by tilting the sample to different angles. The ROIs of each dataset Ti are different. The training dataset of the baseline UNet (B), the baseline UNet initialized with pre-trained parameters from the synthetic dataset (B_SyntExp), the deeply supervised UNet (DS) and the Holistically-Nested Edge Detection (HED) models was composed of 4 images of (Di)i=1,...,4 datasets and one image per (Ti)i=2,...,5 datasets. The two images of the (Di)i=5,6 datasets were used as validation dataset (see subfolder Data/TrainValidData). The entire T1 and T6 datasets were used as testing datasets (see subfolder Data/TestData). A synthetic dataset (see subfolder Data/SyntData where only a few example data are given) was used to compare different parameter initialization strategies. The folder Models contains the trained B, B_SyntExp, DS and HED model parameters and the SEraMic multi-image-based detection results. The j-th detection result of the SEraMic on testing dataset Ti based on n SEM-BSE images is presented by Models/SEraMic/Pred/Ti/Sn/Res_plot/Ij.tif. The set of n images used for one multi-image detection result is given by Models/SEraMic/Multi_image_based_detection.pdf. [1] M. B. Saada, Étude du comportement visco-plastique du dioxyde d’uranium: quantification par analyse EBSD et ECCI des effets liés aux conditions de sollicitation et à la microstructure initiale, Ph.D. thesis, Université de Lorraine, 2017. [2] M. B. Saada, X. Iltis, N. Gey, A. Miard, P. Garcia, N. Maloufi, Influence of intra-granular void distribution on the grain sub-structure of uo2 pellets after high temperature compression tests, Journal of Nuclear Materials 545 (2021) 152632. [3] N. Clavier, M. Massonnet, L. Claparede, R. Podor, P.-H. Imbert, J. Martinez, N. Dacheux, Impact of sintering parameters on the microstructure of homogeneous u1-xcexo2+ 𝛿 ceramics, Journal of the American Ceramic Society (2025) e20376.