Detection of water bodies in hyperspectral images using spectral differential similarity and neural networksrential similarity and neural networks

Abstract

Introduction (I): Efficient material detection in hyperspectral images remains a challenge due to their high dimensionality. This study proposes and evaluates two computational methods—Differential Spectral Similarity (SDS) and Sequential Neural Networks for detecting water bodies in hyperspectral images with 380 reflectance bands. Methodology (M): An adaptation of the CRISP-DM methodology was used in four phases: business and data understanding, data preparation, modeling and evaluation, and method deployment. A dataset of 200 spectral signatures was used to train the models, which were applied to a hyperspectral image of the Manga neighborhood in Cartagena, Colombia. Results (R): Both methods identified water body pixels with precision rates between 13.734% and 16.083%. The SDS method produced fewer detection errors and proved to be, on average, 16.1 times faster in terms of computational performance. Conclusion (C): Given their accuracy and lower complexity compared to more advanced approaches, the evaluated methods—especially SDS—are viable options for integration into real-time environmental monitoring systems.