面向对象的最优分割尺度下多层次森林植被分类

Object-based Multi-level Classification of Forest Vegetation on Optimal Segmentation Scale

对福建省将乐国有林场地物进行分析,结合QuickBird高分辨率影像及其重采样低分辨率影像进行多层次分割,同时结合ESP分割尺度评价工具获取最优分割尺度参数,以实现影像地物最佳分割效果。对不同分割层次下对象光谱特征、纹理特征等进行提取,应用专家知识及统计特征选取影像对象特征和确定阈值,实现研究区域森林地物及树种的分类。对森林植被等分类总精度达到80.53%。

An investigation was conducted by analyzing the groundcover of Jiangle Forest Farm in Fujian Province, and muhi- level segmentation was conducted based on QuickBird high spatial resolution image and its resampling multi-resolution ima- ges by using the ESP （ Estimating the Scale Parameter） to acquire optimal segmentation parameters for different groundeov- ers. The segmentations achieved good results. With the extraction of spectral, textural features, and confirming the threshold by expert knowledge and statistical features, the forest groundcover classification was accomplished with the total accuracy of 80.53%.