一种遥感影像自适应分割尺度的分类方法

Discussion on classification method of adaptive scale based on remote sensing image

针对影像分割中的尺度选取问题,该文基于GF-2多光谱数据提出一种基于自上而下自适应分割尺度的分类方法,该方法在提取每一分割对象光谱、纹理特征的基础上,构建其在各波段复杂度函数,根据每类地物在各波段的复杂度阈值和分类规则,经迭代计算,确定每一对象的最适宜尺度和所属地类,进而得到具有最佳尺度的分割和分类结果。将其与采用ESP尺度分析算法得到的单一最优尺度下的分类结果进行对比分析,结果表明:该方法能够获取与地面目标相匹配的分割尺度,改善了分割效果,提高了分类精度,具有一定实用价值。

Scale segmentation is the key step of object-oriented analysis technology.The quality of which affects the object-oriented classification.However,selecting appropriate scale is still a challenge in multi-scale segmentation.A top-down adaptive scale classification method based on GF-2 multispectral data is proposed.Based on extracting the image features of each segmented object,the complexity index function is constructed,and by setting the threshold and classification rules,the optimal scale and feature type of each segmentation object are determined by the iterative calculation,and the segmentation and classification results with the best scale is obtained.Compared with the segmentation and classification results of single optimal scale obtained by using the ESP optimal scale analysis algorithm,the proposed method can accurately obtain the segmentation scale matching with the ground targets,and greatly improves the accuracy of the segmentation and classification,it is of practical value.