为了考察高光谱技术对海洋溢油种类的鉴别能力, 进行了模拟溢油实验。利用紫外三种波长光源(254nm、302nm、365nm)和可见光两种光源(日光灯和阳光)照射海水上的油墨, 使用AvaSpec-2048地物光谱仪采集5个油种(汽油、柴油、煤油、机油、花生油)的高光谱数据(200～1160nm), 基于这些数据, 通过PCA进行特征优化, 使用前10个主分量构建了支持向量机的油种识别模型, 使用网格法对支持向量机参数C和gamma进行参数寻优, 通过5折交叉验证法进行了结果测试。结果表明, 阳光下的光谱识别率最高, 日光灯下的识别效果最差, 三种紫外灯在波长302nm透射下识别率较高, 在254nm和365nm下的识别效果相当。由此可见, 在油种鉴别过程中光源的类型、照射方式、光强都会影响到油种的识别效果。

In order to investigate hyperspectral technology to marine oil spill ability to identify species, the oil spill experiments are simulated. The ink on sea is exposed using three wavelengths of UV light (254 nm, 302 nm, 365 nm) and visible two light sources (fluorescent and sun), five oil types (petrol, diesel, kerosene, oil, peanut oil) hyperspectral data (200~1160nm) are collected using AvaSpec-2048 field spectrometer, based on these data, the feature is optimized by PCA. Using the first 10 principal components, oil seed recognition model is constructed to support vector machine, the optimization is carried out using the grid method of support vector machine parameters C and gamma parameter, the results of tests are carried out using half of the cross-validation method. Spectrum recognition rate under the sun is highest, the spectrum recognition rate under the fluorescent is worst, three transmission 302nm UV lamp has higher recognition rate, the recognition effect of 254nm and 365nm is under rather. The type of light sources in the oil identification process, exposure mode, the light intensity will affect the recognition results of oil seed.