结合光谱相似和相位一致的高分辨率影像分类

A high resolution image classification method considering spectral similarity and phase consistency

针对面向对象分类结果存在"平滑地物细节"的问题,该文提出顾及光谱相似性和相位一致的高分辨率影像分类方法。该方法首先采用顾及光谱相似性的相位一致的模型方法来获得边缘相应幅度;再采用自动标记分水岭算法对影像进行初分割;顾及相邻分割对象的空间位置、形状、面积等特征的多重约束,提出相邻分割对象合并代价函数模型,对分割结果进行优化;最后结合支持向量机(SVM)对分割对象进行分类。结果表明,本文方法在提高高分辨率影像分类精度的同时,还能保持地物细节。

In view of the existence of＂smooth surface detail＂problem in object-oriented classification results,a high resolution image classification method is proposed by taking into account the spectral similarity and phase congruency.Firstly,aphase-consistent model with spectral similarity is used to obtain the corresponding edge amplitude,and then the automatic marking watershed algorithm is used to complete the high-resolution remote sensing image segmentation,taking into account the multi-constraint of the location,shape,and area of the adjacent segmentation object,the merging cost function model is proposed,and the segmentation result is optimized,finally,the support vector machine（SVM）is used to classify.The experimental results showed that the classification method not only improved the accuracy of high resolution image classification but also kept the details of the objects.