一种面向对象的最优分割尺度计算模型

作为信息提取和分类的前提，面向对象的影像分割尺度参数的设置直接影响到提取和分类的精度。本文以GF-2影像数据为例，在已有分割理论和方法的基础上提出一种基于最优分割尺度的计算模型（OS模型）。该模型以主成分分析所得的主成分以及新建的归一化植被指数（normalized vegetation index，NDVI）特征层作为分割参考层，综合考虑均质因子的影响，构建加权尺度评价指数，插值拟合最优分割尺度。构建误差系数（E_c）对模型进行评价，结果表明：OS模型误差系数（E_c=1.15%）小于传统模型（E_c=3.28%），且分割对象更均匀、与实际地物更接近。

Object-Oriented Optimal Segmentation Scale Calculation Model

Object-oriented image segmentation is the premise of information extraction and classification, and its scale parameter setting directly affects the accuracy of extraction and classification. Taking GF-2 image data as an example, this paper presented a new optimal scale model based on the existing segmentation theory and method. By taking the obtained components of the principal component analysis and the newly built NDVI feature layer as the segmentation reference layers, the authors carried out multi-scale segmentation. In consideration with the influence of shape factor and compactness factor comprehensively, the weighted scale assessment index was constructed, and the cubic spline interpolation was used to fit the optimal segmentation scale. Finally the error coefficient (E_c) was proposed to compare the new model with the original model. The results show that the error coefficient of the OS model (E_c=1.15%) is smaller than that of the original model (E_c=3.28%), and the segmentation objects of the OS model are closer to the ground truth. This model provides an objective basis for the setting of scale parameters, avoids the subjectivity of traditional parameter selection, and improves the image segmentation quality.