基于面向对象的无人机正射影像地物分类

UAV Ortho-Images Classification Based on Object

摘要: 无人机航拍能够快速准确获取地表的高分辨率影像，已经成为遥感数据获取的重要手段之一。采用eCognition软件面向对象分类方法，对无人机影像进行地物分类研究。通过ENVI OneButton生成无人机正射镶嵌影像，选择合适的分割参数对实验区影像进行多尺度分割，找出最优的分割尺度。利用eCognition特征优化功能选择最优对象特征组合，进行最近邻分类。结果表明，分类的总体精度达到83%，Kappa达到0.8，采用eCognition面向对象的分类方法能够较为准确地得到地物覆盖信息。利用无人机技术和eCognition面向对象分类方法，可充分利用影像的光谱信息和形状、纹理等空间信息，能够实现地物信息的快速、准确提取。

Abstract: Unmanned aerial vehicle can obtain high-resolution images quickly and accurately, which had become one of the most important means of remote sensing data acquisition. In this paper, ob-ject-oriented method of eCognition software is used to UAV ortho-images classification. ENVI OneButton was used to generate UAV orthographic mosaic image. We selected the appropriate multi-resolution segmentation parameters for image segmentation and optimal object feature combination using optimization function of eCognition software. Finally, the nearest neighbor method is used for classification. The results showed that the overall accuracy of the classification was 83%, and the Kappa reached 0.8. The objected-oriented classification method of eCognition software can obtain more accurate coverage information of ground objects. combined with UAV technology and objected-oriented classification method, the surface information can be acquired accurately by full use of the spectral ,shape, texture and other spatial information.