Extraction And Analysis Of Oil Film On Water Using Hyperspectral Characteristics

The relative thickness and distribution of the oil film are important parameters, which determine the optimal strategies during assessment and clean-up of the marine oil spills. As a macroscopic observation method, remote sensing plays an increasingly sig-nificant role in the oil spill accident, especially in large-scale oil spills. It has irreplacea-ble advantages over other means.Over the past several years, the airborne and spaceborne hyperspectral remote-sensing has made great achievements with the development of sensor and data processing technologies. The continuous spectral bands make it possible to distinguish the objects with similar spectrum, so as to provide more accurate information of the oil spill. Current researches focus on the spectral characteristics or imagery feature extrac-tion individually, but do not analyze both of them at the same time. In terms of spectral characteristics analysis, most of the researches showed the changes of the reflectance values with different thickness, weathering phases, etc. Few of them analyzed the mor-phological characteristics of the spectra. The differences of some objects at certain bands were used to evaluate the potential application of hyperspectral data directly, which ignored the impact of environmental noise.In this paper, the spectral characteristics of oil film on water with varied thick-nesses and phases were investigated and the coupling relationship among them were studied using wavelet transform. The results indicated the maximum thickness that could be discriminated by optical light was300μm, and the changes of spectrum oc-curred in the first7days. Base on the calculation of the environmental noise equivalent radiance of Hyperion data, the reflectance data of water, light diesel and crude oil were filtered by the response function, and the potential of the Hyperion data to identify the oil film and discriminate different oil type and thicknesses was assessed, which showed that the Hyperion data could discriminate crude oil film with varied thicknesses and the bands between the12th and40th were more effective to identify light diesel oil film. A spectral characteristics based decision-tree classification method was proposed to ex-tract oil spill information from multi-spectral imagery. The classification could be processed by analyzing the spectral characteristics and the image features of oil spill, the accuracy of which was93.7%and satisfied the requirement completely. Aiming at the numerous bands of the hyperspectral remote sensing data, the Minimum Noise Frac-tion(MNF) based decision-tree classification was used to extract oil spill information. This method reduced the data dimension and noise effectively, and magnified the dif-ferences between sea water and oil film, which could get a more accurate and rapid re-sult.