Advanced Segmentation and Geometric Analysis of Geological Samples

The identification of lithology from drill cuttings is a crucial step in reservoir characterization, providing di- rect insights into the drilled rock formations; neverthe- less, traditional cuttings analysis is labor- intensive, re- quires expert interpretation, and often suffers from in- consistency due to its subjective nature. To address these challenges, we propose an automatic, real-time method for drill cuttings characterization. We propound a cut- ting instance segmentation workflow for analyzing 2D images of densely clustered cuttings. Traditional seg- mentation methods struggle to deliver satisfactory results in such complex scenarios. To overcome these limita- tions, we combine marker-based segmentation with a su- pervised deep learning model; nonetheless, training such a model depends on the quality of the reference dataset, and annotation is typically time-consuming and costly. The novelty of our approach lies in its fully automated workflow, eliminating the need for manual annotation or refinement. We validate our approach on a large, diverse dataset of real drill cuttings, demonstrating high performance across various lithologies and cuttings’ sizes. Index Terms— Instance segmentation, Geometric analysis, drill cuttings, convolutional neural networks (CNN) 1