Superpixel segmentations for thin sections: Evaluation of methods to enable the generation of machine learning training data sets

Abstract

The lack of well-labeled training image data has significantly impeded the development of novel DL methods in mineral thin section images identification. However, image annotation, especially pixel-wise annotation is always a costly process. Manually creating dense semantic labels for rock thin section images has been long considered as an unprecedented challenge in view of the ubiquitous variety and complexity of minerals in thin sections. To speed up the annotation, we propose a human–computer collaborative pipeline in which superpixel segmentation is used as a boundary extractor to avoid hand delineation of instances boundaries. Thin section data sets also pose specific requirements to superpixel segmentation algorithms. The most obvious aspect is that thin section data sets contain more than just a single image, due to the combined use of plane-polarized light views and cross-polarized views at different angles. Due to this limitation, we implemented the extension of an existing algorithm, SLIC, to use multiple image layers, resulting in the adapted algorithm MultiSLIC. We evaluated several algorithms with respect to their use for the specific requirements of thin section data sets, Both qualitative and quantitative evaluation studies show an overall good performance of MultiSLIC in terms of boundary adherence and compactness of the resulting segmentation. This is an important aspect for the subsequent labeling by human annotators with a specifically designed labeling tool. We also provide a prototype of superpixel labeling tool in python.

Jiaxin Yu

My research encompasses both theoretical investigations in the domains of rock and subsurface studies, as well as the practical application of AI technologies to geoscientific endeavors. On the theoretical front, my research interests are primarily centered around rock physics, granular medium theory, and poroelasticity. From an applied perspective, my focus lies in leveraging artificial intelligence for rock mineral identification, employing Graph Neural Networks (GNNs) for simulating granular mediums, and contributing to open-source initiatives dedicated to rock physics and computational geoscience projects