A FIELD EXPERIMENTAL APPROACH TO HYPERSPECTRAL REMOTE SENSING FOR XYLENE AND TOLUENE SPILL MONITORING

Abstract

The burgeoning chemical industry has precipitated a significant escalation in the transportation of hazardous and noxious substances (HNS), inherently increasing the risk of maritime spill incidents due to the reliance on sea freight. Marine HNS spills pose significant fire and explosion hazards, necessitating the development of advanced remote sensing technologies for wide-area, high-resolution detection. This study presents detection results from a ground experiment using xylene and toluene spills in an outdoor marine pool. The hyperspectral sensor was mounted at a height of approximately 12 m using a lift, allowing for vertical observation of the ground pool. For hyperspectral HNS detection, principal component analysis (PCA) was applied to reduce dimensionality. The N-FINDR technique was tested as an HNS detection algorithm, but faced challenges in distinguishing between HNS and seawater due to similarities in endmember characteristics. Among the methods tested, the K-means algorithm demonstrated superior performance in detecting HNS and effectively identified bubbles within the HNS. To implement the K-means algorithm, it was necessary to predetermine the number of clusters. the decision was made to allocate either 2 or 3 clusters based on the inclusion of bubbles as a separate category.