讨论了城镇绿地树种识别数学描述符的设计思想和方法。本着具有确切的物理意义、几何意义或植物生态 学意义以及分割阈值具有环境不变性的原则，设计了归一化阴影指数、饱和度明度相对差、相对边缘点数、相对暗 细节密度、相对骨架密度和加权平均冠径等14个分别涉及波谱、纹理和形状特征的新描述符。经过样本统计分析和 遥感图像实例测试，证明这些描述符在城镇绿地树种识别方面比经典描述符具有更好的针对性和更强的适应性。此 外，本文还讨论并测试了红色欠饱和像元补偿集的提取方法，以及基于cell分割或分类的方法。对于城镇绿地树种 分类问题，在决策树分类输入矢量中，使用本文的描述符组合误分率为5.8%，相比传统的分色亮度组合(误分率为 25.9%)有明显改进。

This manuscript has the merits of providing a useful means to identify plant species of urban landscape vegetation from high-resolution remote sensing images. The study designed and selectively tested an array of quantitative descriptors calculated using spectral, textural, and shape characteristics of image objects. These descriptors, theoretically independent of image types and acquisition environment, may signifi cantly improve the capacity of machine learning and discrimination of some classifi ers. The demo cases indicated that with a combination of four such descriptors to identify plant species, the error rate is no more than 5.8% while comparing 25.9% with the conventional spectrum-based approach.