面向对象矿区复垦植被分类最优分割尺度研究

发布时间：2018-05-03 10:27

本文选题：复垦植被 + Worldview-影像　；参考：《测绘科学》2016年04期

【摘要】：针对面向对象分类中直接影响最终分类精度的分割尺度问题,该文通过基于对象的精度评价,定量对比分析eCognition软件自带的ESP工具、欧几里得距离二指数(ED2)和分割误差(SE)3种分割尺度选择方法,研究不同尺度在矿区复垦植被面向对象分类的精度差异,为高分遥感数据分割尺度的选择提供定量的标准。得出结论:分割尺度的微小变化会导致分类结果的显著差异;ESP工具分类精度最低,SE和ED2全面考虑参照样本与分割对象之间的特征和几何差异,可显著提高面向对象分类精度;将SE和ED2作为最优分割尺度选择算法嵌入面向对象分类体系中,可以快速获取最优分割尺度。
[Abstract]:Aiming at the segmentation scale problem which directly affects the final classification accuracy in object-oriented classification, this paper compares and analyzes the ESP tool of eCognition software through the object based accuracy evaluation. Euclidean distance (Euclidean distance 2 index ED2) and segmentation error are used to study the accuracy differences of different scales in the object oriented classification of reclaimed vegetation in mining areas, and to provide a quantitative criterion for the selection of segmentation scale for high score remote sensing data. It is concluded that the small variation of segmentation scale will lead to significant differences in classification results. ESP and ED2 have the lowest classification accuracy and consider the characteristics and geometric differences between reference samples and segmented objects in a comprehensive way, which can significantly improve the accuracy of object-oriented classification. Using SE and ED2 as the optimal segmentation scale selection algorithms, the optimal segmentation scale can be obtained quickly by embedding them into the object-oriented classification system.
【作者单位】：东北大学资源与土木工程学院;中国矿业大学感知矿山(物联网)中心;辽宁国土资源规划院辽阳分院;
【基金】：中央高校基本科研业务经费项目(N120801002) 国家自然科学基金项目(4140010440)
【分类号】：P237

【相似文献】