Hyperspectral remote sensing and clustering techniques for monitoring marine hazardous and noxious substance spills

Abstract

Marine hazardous and noxious substances (HNS) spills pose severe environmental and safety risks, including fire, explosion, and toxicity, necessitating the advancement of high-resolution remote sensing technologies for effective detection and monitoring. However, research in HNS monitoring remains limited and underexplored, with significant gaps in the application of remote sensing techniques. This study investigates the potential of hyperspectral imaging for HNS spill detection through controlled field experiments. Xylene and toluene were introduced into an outdoor marine pool, and a visible near-infrared (VNIR) hyperspectral sensor was deployed to capture their spectral characteristics. Unsupervised clustering techniques, including K-means, and Gaussian mixture models (GMM), were applied following dimensionality reduction via principal component analysis (PCA). While N-finder algorithm (N-FINDR) struggled to differentiate HNS from seawater dueto spectral similarities, the K-means algorithm demonstrated superior performance, effectively classifying HNS and detecting coexisting bubbles. The GMM approach further validated the classification accuracy. These findings highlight the feasibility of clustering-based hyperspectral detection for marine HNS spills, offering a promising framework for enhancing emergency response and environmental monitoring.