A region-growing approach for spectral bandprioritization in hyperspectral remote sensing

Hyperspectral imaging (HSI) is increasingly applied in remote sensing, generatinglarge volumes of high-dimensional data that require advanced analysis methods.Traditional approaches often rely on large amounts of manually labeled dataor contextual information and tend to focus primarily on the spatial dimen-sion, underutilizing the rich spectral content of HSI. This study introduces anovel method for identifying informative spectral bands using a 3D region grow-ing algorithm (3D RGA), which enables segmentation based on both spatialand spectral information. The proposed approach allows for the preprocessing ofunlabeled HSI datasets by reducing the number of spectral bands while main-taining high segmentation accuracy. It also enables the detection of anomalieswithin labeled regions, offering insight into the physical characteristics of differ-ent land cover types. Through both qualitative and quantitative evaluations onpublicly available and real-world datasets, the method demonstrates its abilityto isolate distinctive spectral profiles, identify relevant bands for specific classes,and support cross-dataset generalization. This work presents an interpretabletechnique for spectral band selection that enhances segmentation, facilitatesclass-level analysis, and lays the groundwork for future applications in automatedannotation and monitoring.