4D电子衍射数据集

本研究使用名为‘4D电子衍射数据集’的大型数据集，由材料科学电子显微镜(EMAT)和NANOlab卓越中心创建。该数据集包含约126000个样本，主要来源于Materials Project的晶体数据文件。数据集的创建涉及使用MULTEM软件模拟多切片算法和显微镜模型，模拟参数和显微镜设置随机抽取自实际可行的范围。数据集主要用于训练深度学习模型，解决从噪声/稀疏强度记录中检索复杂波的问题，并进一步从卷积神经网络(CNN)预测的出口波中重建样品图像。该数据集的应用领域主要集中在低剂量范围内的2维和薄样品的高分辨率成像，旨在通过深度学习技术提高电子显微镜数据的解析度和灵敏度，特别是在低剂量实验中。

This study employs a large-scale dataset titled the '4D Electron Diffraction Dataset', which was developed by the Materials Science Electron Microscopy (EMAT) and the NANOlab Center of Excellence. This dataset contains approximately 126,000 samples, primarily derived from crystal data files of the Materials Project database. The development of this dataset utilized the MULTEM software to simulate multi-slice algorithms and microscopy models, with simulation parameters and microscope configuration settings randomly sampled from empirically feasible ranges. This dataset is primarily used for training deep learning models to address the challenge of retrieving complex wave functions from noisy or sparsely sampled intensity recordings, and further reconstructing sample images from exit waves predicted by convolutional neural networks (CNNs). Its core application scenarios focus on high-resolution imaging of 2D and thin specimens under low-dose conditions, aiming to improve the resolution and sensitivity of electron microscopy data via deep learning technologies, particularly in low-dose experimental settings.