无人机影像分割尺度及地物分类研究

Research on Segmentation Scale of UAV Image and Ground Feature Classification

本文对面向对象的遥感图像分类技术及影像分割方法进行了介绍,重点对影像分割准则、多尺度分割及最佳分割尺度选择进行深入分析,建立了基于分割方法原理和影像特征的最佳分割尺度选择策略。以无人机高分辨率遥感影像为数据源,对多尺度分割及最佳分割尺度的选择进行实验研究,对比分析了不同尺度下的分割效果,并据此确定了最佳分割尺度。最后基于面向对象的分类方法实现了实验区地物分类和提取,并对分类结果进行了精度评定。结果表明:通过选择最佳的分割尺度,可获取较高的地物分类精度,且可以避免传统分类方法中的"椒盐"现象。

This paper introduces the object-oriented remote sensing image classification technique and image segmentation, with emphasis on image segmentation criterion, multi-scale segmentation and best-fit segmentation scale selection, establishing the best-fit segmentation scale selection strategy based on the segmentation principle and image features. By carry on experimental studies of multi-scale segmentation and best-fit segmentation scale selection based on the data of UAV images, we compare the effects under different scales, and find out the best-fit scale. Finally, based on the object-oriented classification method, we can classify and extract the surface features in the study area, and assess its accuracy. The results demonstrate that under the best-fit segmentation scale, we can obtain better classification accuracy and avoid the ＂salt and pepper＂ phenomenon at the same time.