Probing Volcanic Systems with Machine Learning for Quantitative Petrography

该项目的重点是开发机器学习技术，用于从光学和电子显微镜中量化火成岩的性质。在我们最近成功地将SEM图转换为冰岛巨大的拉基火山喷发的纹理特性和结晶时间尺度的量化之后（Neave等人，2017年），我们计划进一步探索机器学习在火成岩图像分析中的可能性（Andrew， 2018年-蔡司的合作伙伴之一）。我们最近还展示了电子后向散射衍射（EBSD）分析火山晶体的能力（Wieser等人，2020），并旨在实现这种新方法的潜力。蔡司的合作伙伴也在探索在地质应用中融合3D和2D成像，这也可能是研究的重点。该学生将获取并处理冰岛火山岩的光学和电子显微镜图像。学生也可能希望在冰岛进行实地采样，以获得适合图像分析的样本（例如，具有玻璃状地面质量的糊状结节）。然后，他们将开发图像处理技术，包括机器学习，以自动识别相位和晶体分割。然后，他们将开发进一步的处理步骤，从分割图像中提取有用的纹理信息：晶体尺寸和形状分布、聚类分析、接触角的量化、内部变形的表征。然后将这些特性与自然和实验数据的参数化进行比较，以提取有关物理过程的信息（例如，岩浆在活火山下上升的时间尺度）。

This project centres on the development of machine learning techniques for the quantification of igneous rock properties from optical and electron microscopy. Following on from our recent successes in conversion of SEM maps into textural properties and quantification of a crystallisation timescale for the gigantic Laki eruption in Iceland (Neave et al., 2017), we plan to further explore the possibilities for machine learning in image analysis of igneous rocks (Andrew, 2018 - one of our partners from Zeiss). We have also recently demonstrated the power of Electron Backscatter Diffraction (EBSD) analysis of volcanic crystals (Wieser et al., 2020) and aim to realise the potential of this novel approach. Partners in Zeiss are also exploring merging 3D and 2D imaging in geological applications, and this may also be a focus of study.The student will acquire and process co-registered optical and electron microscope images of volcanic rocks from Iceland. It may also be desirable for the student to carry out field sampling in Iceland to obtain suitable samples for image analysis (e.g. mushy nodules with glassy ground mass). They will then develop image processing techniques, including machine learning, to automate phase identification and crystal segmentation. They will then develop further processing steps to extract useful textural information from the segmented images: crystal size and shape distributions, clustering analysis, quantification of contact angles, characterisation of internal deformation. These properties will then be compared with parameterisations of natural and experimental data to extract information about the physical processes involved (e.g. timescales magma rise under active volcanoes).