Analysis of the performances of hyperspectral lidar for water pollution diagnostics

The paper is aimed at the analysis of the performances of hyperspectral lidar for detection and classification of oil pollution in water in comparison with a laser fluorosensor having a few discrete detection channels only. It is demonstrated that hyperspectral laser-induced fluorescence (HLIF) spectra include all relevant spectral information about the target in contrast to discrete detection channel sensor data. In order to extract significant features from HLIF data, a multi-resolutional analysis, namely the discrete wavelet transform (DWT), is applied. The feature extraction is automated using the sparsity-norm optimization method. The resulting features have a clear spectral representation and are used in automatic object classification. The classification results and selectivity are compared with discrete detection channel sensor data on a number of oil pollutants. The results of simulation experiments demonstrate the high value of classification accuracy and the ability to sub-classify similar organic compounds from single groups of objects. A comparison with discrete channel sensor data shows a significant increase in the overall performance of oil pollution detection and classification.