基于多尺度光谱角制图的遥感影像单木树冠提取方法

Multi-scale Spectral Angle Mapper Method for Individual Tree Crown Delineation from Remote Sensing Image

针对现有的单木树冠提取方法在尺度直径丰富的林区提取精度低、过分割和欠分割现象严重的问题,提出了一种改进的基于光谱角制图的多尺度层级化标记控制点的分水岭分割方法。首先通过形态学变换得到形态学梯度图像;接着通过HSV变换得到亮度图像,之后用高斯滤波去除噪声,使用区域最大值法在多个不同尺度上识别树冠标记,利用同心叠加整合不同尺度对象上的标记,并用光谱角制图控制叠加过程形成最终标记;最后实施标记控制的分水岭分割得到单木树冠分割线。以3种不同类型的高分辨率林区作为测试数据,同时采用3种传统单尺度标记控制的分水岭分割算法作为对比试验,结果表明,该方法对树冠的提取精度均在80%以上,且较单尺度的方法而言,能够有效地改善树冠提取中的过分割和欠分割问题。

In order to solve the problems of low precision, severe over-segmentation and under-segmentation in forest areas with abundant size and diameter, an improved marker-controlled watershed segmentation with multi-scale hierarchical marker control points is proposed. Firstly, a morphological gradient image is generated. Then, the brightness image is obtained by HSV transformation, and the noise is removed by Gaussian filtering. The tree crown markers are identified at multiple different scales by using the region maximum method, the concentric superposition is used to integrate the multi-scale markers to form the final marker map, and the process of forming the marker map is controlled by spectral angle mapper(SAM). Finally, a marker-controlled watershed segmentation is implemented. The proposed algorithm, together with other three single scale markers methods are tested using three different high resolution forest regions. The results show that the accuracy of the proposed method is above 80%, and compared with the single-scale method, the proposed method can effectively suppress over-segmentation and reduce under-segmentation of the individual tree crown delineation.