Prediction Model Study Of Oil Thickness Based On Spectral Curve Response Characteristics

Today, Marine oil spill accidents occur frequently. How to estimate the amount of spilled oil is an important topic. If get the oil spill amount, it is helpful to the subsequent processing and loss assessment. Due to hyperspectral develops rapidly, to estimate oil thickness becomes possible.Hyperspectral remote sensing technology is widespreed concerned as a popular remote sensing technology. The main reasons are as follow:hyperspectral remote sensing technology can get the substance data that can not be obtained by normal remote sensing image, and hyperspectral remote sensing technology can link substance spectral data with its attributes. In this way, the analysis of the substance can be seen as dealing with the difference and similarity of spectral curves. The premise of which is that hyperspectral technology can be used for estimating film thickness.First, measure the different oil thickness using AvaSpec Spectrometer and get the corresponding spectral curves. Then analyze the relationship between the oil thickness and the different characteristics of these curves. The study shows that the oil thickness has large correlation with variables based hyperspectral position such as Rg, Ro, and vegetation indexes such as RDVI, TVI, Haboudane.By curve fitting, BP neural network, SVD-based iterative method and Mulit-feature fusion curve fitting, build the predicted relationship between curve characteristics and oil thickness, and using it to predict oil thickness of the different spectral curves of oil. Finally analyze each of the estimated models by comparing time and precision.Finally, the methods and models use in the real hyperspectral images, to achieve forecasts of film thickness.