Evaluation and optimization of the three-band spectral indices for oil type identification using reflection spectrum

Oil type identification can help determine the source of leakage and decide a plan of post-accident treatment. Visible spectroscopy is a promising way for rapid oil type identification. Choosing the appropriate three-band spectral index (TBSI) and corresponding optimal band combination is the essential basis to improve the accuracy of the identification model. In order to determine the TBSIs that are available for oil type identification, as well as the corresponding optimal spectral band combination, this study proposed a novel algorithm for the optimization of spectral band combination for classification problems by evaluating the distinguishability of TBSI with statistical parameters from paired t-test. Heavy diesel, refined diesel, and the crude oil originated from two different places were used in the experiments, in which the spectra of the oil samples were collected using high-resolution hyperspectral sensor. Four different TBSIs that are commonly-used in the reflection spectrum analysis were tested using the proposed optimal band combination algorithm. According to the results, three out of the four TBSIs tested in the experiment are proved to be applicable for oil type identification. The optimal spectral band combinations are determined for each TBSI, and their distribution patterns in the three-dimensional tensor of spectral band combinations are studied and discussed. The optimal spectral combination algorithm proposed in this study can not only determine the spectral bands for rapid detection of oil spills with smaller number of calculations, but also provide the theoretical basis for other target classification tasks using reflection spectrum.