Application of YOLOv11 Deep Learning Model for Classification and Counting Ice-Rafted Debris (IRD) in Core Sediments in the Arctic Ocean, Submitted to Artificial intelligence in geoscience.

This repository contains code and pre-trained models for automated classification and counting of ice-rafted debris (IRD) grains in Arctic Ocean sediment core using a two-stage YOLOv11 deep learning approach. The methodology consists of two sequential stages:1. Stage 1 (Multi-grain Segmentation): YOLOv11x-seg detects and segments all IRD grains in thin section images at 2048px resolution2. Stage 2 (Single-grain Classification): YOLOv11x-cls classifies each cropped grain into four lithological categories: Clastic, Crystalline, Detrital Carbonate, and Quartz This automated workflow significantly reduces the time and expertise required for traditional manual IRD analysis, enabling rapid quantification of IRD provenance and composition in Arctic sediment cores. Package contents:- Pre-trained model weights (Stage 1: 120 MB, Stage 2: 55 MB)- Image preprocessing code (CLAHE and Gamma normalization)- Training scripts for SLURM-based HPC systems- Sample training data (3 multi-grain images, 12 single-grain crops)- Comprehensive documentation and training guide Model Performance:- Stage 1: mAP@50 98.2%, Precision 96.5%, Recall 94.8%- Stage 2: Top-1 Accuracy 87% Published in: Artificial Intelligence in GeosciencesContact: jkeum@gnu.ac.kr