| Summary: | An accurate identification of oil spill types is the basis of determining the source of leakage, evaluating the potential damage, and deciding a plan of responses for an oil spill event. Despite sufficient studies that interpreted and analyzed hyperspectral data of oil spills, these studies that identify or classify oil spill types is rather limited. Aiming at identifying different types of oil spills, this article analyses the reflectance spectra obtained from high-resolution hyperspectral sensors using multiple machine learning methods. Four types of machine learning models are applied in this article: random forest; support vector machine (SVM); and deep neural network (DNN); and DNN with differential pooling (DP-DNN). The training and testing data are collected by field experiments under different environmental condition in order to verify the robustness of the machine learning models. The characteristics of reflectance is briefly described, and the results conform with results from previous studies. The performances of the machine learning models are evaluated and compared in terms of both accuracy of prediction and computational complexity. The results indicate that the two DNN models are able to achieve the most accurate prediction among the four machine learning models at the cost of more computation. The SVM model, or the proposed DP-DNN model may be a favorable choice when training time is limited.
|
|---|
