Determining the Sphericity of Sand Grains by AI Computer Vision Research

Abstract

The paper presents an approachable, free, open source and effective way of using AI models in geotechnics. The research consisted of two stages. First, granulometric tests were performed according to EN ISO 14688–1: 2018. Second, sand grains were identified and measured in gathered microscopic images (from a light microscope and an SEM). Segmentation, which allows the identification of image pixels belonging to an object, is a well-known but problematic area of image processing, and the use of AI techniques is one of the possible approaches to dealing with the issues involved with segmentation. The estimation of grain sphericity is used as an example to present how geotechnics specialists can customize AI and other software for their benefit with only basic knowledge of the Python programming language.