A BOX-YOLOv8 framework for automated hyperspectral core analysis

• A framework for automated core analysis has been proposed by integrating hyperspectral imaging technology with deep learning. • An ensemble of various deep learning models is utilized for core extraction. • BOX-YOLOv8 improves accuracy & speed over YOLOv8m for hyperspectral core analysis. • Spatial transforms integrate data from three diverse cameras for comprehensive borehole analysis. • Framework automates core analysis, accelerating geological assessment & decision-making. Hyperspectral imaging (HSI) offers significant potential for soil and geological core analysis, yet challenges of large data volume and complex processing hinder its broader usage. To tackle these issues, we introduce an innovative automated framework for hyperspectral soil and core analysis. This framework leverages a synergistic integration of deep learning models—specifically BOX-YOLOv8, SAM2, and a custom MLP—to achieve efficient, automated end-to-end processing. The framework identifies core boxes (BOX-YOLOv8, achieving 99.5% [email protected]), precisely segments these boxes for geometric correction (SAM2), and segments core/soil material from the background using spectral differences (MLP). This automated pipeline significantly improves processing time compared to manual methods. A key innovation is the accurate (sub-1-pixel) spatial transformation enabling the fusion of multi-sensor data (SWIR, MWIR, RGB) for comprehensive borehole characterization. Furthermore, the framework facilitates detailed spectral analyses, including mineral mapping and spectral library matching, advancing soil-related analysis by providing insights into composition and structure. Overall, this deep learning-driven framework significantly automates soil/core analysis, improves assessment speed, and enhances decision-making in resource exploration, soil survey, and environmental studies.