【摘要】 针对海上溢油SAR图像中油膜与类油膜的识别问题,提出了一种结合傅立叶分形与特征提取的检测算法。由于分形特征可以具有无穷多的细节,并在不同的研究尺度存在自仿射特性。这与油膜及类油膜表面的几何形貌特征非常吻合。该算法通过计算样本的傅立叶分形特征,组成油膜与类油膜的特征空间。然后,应用基于差分进化的特征选取方法将利于分类的重要特征值筛选出来。再利用重要特征值对原有样本进行分类。实验表明,经特征选取的分形特征向量能够以100%的准确率将两类样本区分开。该算法在选取重要特征的同时实现了对高维特征空间降维的目的,该思想可以应用于其他的基于高维特征的识别系统中,具有普遍的适用性。更多还原

【Abstract】 In this paper,a novel algorithm is present for oil spill detecting from look-alikes on SAR image.It based on Fourier fractal and feature selection.Fractal characteristics have an infinite number of details,and it has self-affine characteristics in different scales.These geometric characteristics are identical with oil film and look-alikes surface very well.The feature space of this method is composed by Fourier fractal characteristics.By using the differential evolution feature selection method,we select important eigen values(IEs)from total sample space.By using ANN and IEs,we classify the sample space.Experimental results show that classify accuracy rate reach the 100% when using selected eigenvectors.The algorithm gets the purpose of dimension reduction while select the important feature.This idea can be applied to other high-dimensional feature-based recognition system.更多还原