海洋溢油给生态系统和环境带来了严重的损害, 及时准确地探测海洋表面的溢油对维护海洋生态平衡和保护环境具有至关重要的意义。目前, 在溢油检测方面, 溢油特征选择方法人为主观性强、特征利用效率低、较少涉及多种不同特征组合的应用。为此, 提出一种基于 RF-RFECV海洋溢油特征智能优选策略。通过系统提取Radarsat-2影像的极化特征与纹理特征, 构建多维特征空间。算法自主计算各个特征在溢油、疑似溢油及海水类别中的区分度权重, 最终为Radarsat-2数据筛选出16/18维最优特征组合。实验结果表明, 利用优选后的 特征进行溢油检测, 有效提升了溢油检测的精度, 降低了误检和漏检的情况, 充分证明了算法的有效性。

Marine oil spills have caused serious damage to the ecosystem and the environment, and timely and accurate detection of oil spills on the ocean surface is of vital significance to the maintenance of the marine ecological balance and the protection of the environment. At present, in terms of oil spill detection, the oil spill feature selection method is artificially subjective, the feature utilisation efficiency is low, and the application of multiple different feature combinations is less involved. For this reason, a RF-RFECV marine oil spill feature intelligent preference strategy based on RF-RFECV is proposed. The polarisation features and texture features of Radarsat-2 images are extracted systematically to construct a multi-dimensional feature space. The algorithm autonomously calculates the differentiation weights of each feature in the categories of oil spill, suspected oil spill and seawater, and finally selects the 16/18-dimensional optimal feature combinations for Radarsat-2 data. The experimental results show that the use of the optimal features for oil spill detection effectively improves the accuracy of oil spill detection, reduces the cases of misdetection and omission, and fully proves the effectiveness of the algorithm.