为利用不同油种的发光特性来探测海洋溢油，通过高光谱成像仪，在两种照明模式下采集了6种溢油油种的高光谱图像。基于33个波段构建了波段均值、波段差、波段比和归一化波段比4个辐射指数，提出了基于Fisher和PCA的模型共识的溢油高光谱特征选择方法，采用RBF-SVM模型对油种进行识别。比较发现，本文构建的基于光源混合、波段运算和模型共识的多模式融合方法，从不同侧面提高了模型的溢油识别能力，识别率达到了99.1%以上，比单一方法提高了10%以上。结果表明，多模式融合有效提高了海洋溢油的识别率。

In order to identity different oil spill by the fluorescence phenomena of oil and its products, the hyperspectral images data of six varieties of oil spills samples were collected under two kind of illuminations (UV and halogen lights) using hyperspectral imaging camera. In the spectral region of 400-720 nm (10 nm spectral bandwidth), four radiation index were obtained which include radiation index of individual spectral bands and the difference, ratio, and the normalized difference radiation index of consecutive spectral bands. Then, a novel method composed of Fisher and PCA to identify most significant wavelengths was proposed, and a classified model based on REF-SVM and the proposed method was established. By comparison, it is found that the different radiation index, light fusions and model consensus of feather selected method all can improve the accuracy of recognition rate. The overall accuracy rate by our method is above 99.1%, which is obviously higher than traditional methods only use one method. The experiment results show that the multi-pattern fusion can effectively improve the recognition rate of marine oil spill.