基于改进分割算法的退耕地树冠信息提取

Tree-crown information extraction in returning farmland to forest land based on improved segmentation algorithm

优化特征空间和改进分割算法是基于面向对象技术实现退耕地树冠信息准确提取的重要环节,同时也是高分影像地物识别研究中需迫切解决的问题。为此,以北京张山营镇部分区域的Quick Bird影像为数据源,根据光谱阈值实现一级分割得到林地区域,同时采用区域进化的区域增长算法对改进均值滤波算法去噪声处理后的全色波段执行二级分割,最后结合形状、光谱、纹理指标构建的特征空间完成了树冠信息提取。结果表明:改进分割算法总体精度达91.5%,Kappa系数为0.836 2,分别较传统方法提高了15.5%和0.120 4。

Optimization of feature space and improvement of segmentation algorithm are the important links of accurate tree-crown information extraction in returning farmland to forest land based on the object-oriented technology. They are also an urgent task to solve the problem on the research of feature recognition by high resolution image. QuickBird image of parts of Zhangshanying in Beijing was selected as data source in this study. Firstly, forest area was obtained by the first level segmentation according to spectral threshold. Secondly, the second level segmentation was performed on the panchromatic band dealt with the noise by improved median filter algorithm according to the regional growing algorithm based on regional evolution. Finally, tree-crown information extraction was completed using the feature space built by shape, spectral and texture index. Results showed that the overall accuracy of the method in this paper was 91.5% and the KAPPA coefficient was 0. 836 2, increased by 15.5% and 0. 120 4 than that of the traditional method respectively.