Marine Hazardous and Noxious Substances Spill Monitoring: An Experimental Approach to Hyperspectral Remote Sensing and Clustering Techniques

Abstract

Marine hazardous and noxious substances (HNS) spills present significant environmental and safety hazards, necessitating advanced remote sensing methodologies for their detection. This study explores the potential of hyperspectral imaging for HNS spill monitoring through a series of controlled experiments. Specifically, xylene and toluene were spilled into a marine pool, and a visible near-infrared (VNIR) hyperspectral sensor acquired their spectral characteristics. principal com ponent analysis (PCA)-based dimensionality reduction was followed by unsupervised clustering using N-finder algorithm (N-FINDR), k-means, and Gaussian mixture model (GMM). While N-FINDR encountered challenges in differentiating spectra, K-means demonstrated efficacy in classifying HNS and detecting bubbles, a finding that was further validated by GMM. The findings of this study substantiate the viability of clustering-based hyperspectral detection for enhancing HNS spill monitoring and emergency response.