A configurable five-step workflow for high-throughput hyperspectral image analysis and vegetation index mapping

This workflow describes a structured approach for hyperspectral image analysis of push-broom camera data acquired under laboratory and field conditions. The workflow comprises five sequential steps: (1) ROI extraction from reflectance-corrected hyperspectral datacubes; (2) unsupervised spectral clustering via UMAP dimensionality reduction and DBSCAN; (3) supervised pixel-level classification using a Random Forest Classifier (RFC) with highlight mask generation; (4) computation of vegetation indices from configurable formula definitions with summary outputs; and (5) frequency distribution analysis across experimental treatments, including pairwise histogram overlay for visual comparison. Analysis parameters are configurable so that the workflow can be calibrated to specific imaging conditions. The workflow was demonstrated using wheat-kernel and turfgrass datasets spanning VNIR and SWIR spectral ranges under indoor and outdoor imaging conditions. A five-step workflow with user-adjustable parameters adaptable to diverse hyperspectral imaging conditions. An unsupervised-supervised spectral classification pipeline (UMAP → DBSCAN → Random Forest Classifier (RFC)) for pixel-level segmentation. An integrated pipeline from ROI extraction to treatment-level frequency distribution comparison with numerical histogram output.