面向对象的PolSAR图像SVM分类

Object-oriented SVM classification of PolSAR image

极化合成孔径雷达(PolSAR,polarimetric synthetic aperture radar)图像具有强相干斑噪声和大场景特点,为此提出一种面向对象的支持向量机(SVM, support vector machine)分类算法。算法首先通过超像素分割产生待分类对象,以此减少分类处理单元,同时实现特征滤波降噪;然后通过转换矩阵提取信息完备且具有简单统计描述的雷达散射截面积特征;最后,选择在小样本条件下仍具有较强学习能力和泛化能力的SVM分类器实现图像分类。用公开的实测San Francisco数据进行实验,实验结果表明:该算法相对于对比算法在准确率上提升约10%。

Polarimetric synthetic aperture radar(PolSAR) image is characterized with strong speckle noise and big scene.Therefore, an object-oriented support vector machine(SVM) classification algorithm was proposed. Firstly, the object to be classified is generated by superpixel segmentation, which reduces the classification unit and achieves noise reduction;Then, the radar cross section(RCS) features with complete information and simple statistical description are extracted by the conversion matrix;Finally, SVM classifier with strong learning ability and generalization ability is selected to realize image classification. Experiments are conducted with publicly measured San Francisco data, and the experimental results show that the accuracy of this algorithm is improved by 10% compared with the contrast algorithm.