Instance segmentation model applied to automated palynofacies area-ratio quantification analysis for black shales of T-OAE

For the purpose of this study, an instance segmentation pipeline based on the YOLOv26 model was used for automated detection, classification, visualization, and quantitative analysis of palynofacies in black shales from the Toarcian Oceanic Anoxic Event (T-OAE). With the help of the dataset which contained 564 microscopic images and 2964 annotations in four classes amorphous organic matter (AOM), phytoclasts, palynomorphs, and background, the model achieved a mean average precision at 0.5 IoU threshold ([email protected]) of 94.1% on segmentation masks, with aggregate precision of 93.1% and recall of 92.1%. At the individual class level, the model achieved a mAP of 98.7% for palynomorphs. The mAP of palynomorphs was 98.7%.Beyond object detection, the system automates palynofacies analysis by combining model predictions with a custom built Python based analytical engine. This engine algorithmically extracts Hue, Saturation, and Value (HSV) color features from segmented particles to determine an objective Thermal Alteration Index (TAI) maturity score. Simultaneously, the system calculates the relative surface area percentages of each organic component to generate automated Tyson ternary plot diagrams. This end to end pipeline converts raw optical microscopy images into high-throughput, reproducible quantitative data, significantly accelerating palynological workflow times while eliminating subjective visual estimation bias.