利用新特征空间的SAS图像目标分类算法

摘要

An algorithm of SAS image object classification using a new feature space is proposed on the basis of analyzing dif-ference of statistical features. In the proposed algorithm, Markov random field is used to segment the object and shadow from background, and then the shadow geometrical features and object central moments are computed. Moreover, the differences of statistical parameters of every part are estimated to make up a new feature space with the two mentioned previously. k-mean clus-tered algorithm is applied to classification. The validity of the proposed algorithm is proven by SAS lake-trial.%通过分析合成孔径声纳图像中不同目标统计特性参数间的差异，提出了一种利用新特征空间的SAS图像目标分类算法。该算法用马尔可夫随机场分割算法找到感兴趣区域，提取阴影的几何参数和目标的归一化中心矩，并且将目标、阴影、背景之间统计特性的分布参数之差与前两者构成新的特征空间。利用k-均值聚类算法对三类目标进行分类。合成孔径声纳湖试数据验证了算法的有效性。