基于灰度梯度图像分割的单木树冠提取研究

发布时间：2019-04-24 20:08

【摘要】：树冠是树木的重要组成部分,基于遥感影像的树冠提取对于森林资源调查监测具有重要意义,但准确获得树冠的形状和边界比较困难。高分辨率影像具有丰富的纹理和光谱信息,基于高分辨率影像单木树冠勾勒技术为森林资源调查提供了一种快速有效的测树途径。但是,由于高分影像信息冗杂,面向对象的分割方法数据计算量大,并且需要人工设置光谱或纹理阈值才可以实现单木分割,导致工作效率下降,鲁棒性差。图像增强通过改变原始图像的结构关系,有选择地突出或者抑制图像中的某些特征,有效的图像增强有益于提高单木树冠分割的准确程度。因此,本文提出一种基于影像的灰度梯度图像分割的树冠提取方法,通过对比传统的罗伯斯、拉普拉斯算子与改进的数学形态学算子,利用目视解译与灰度直方图结合的方法确定最优选择为改进的数学形态学算子。然后,利用改进的数学形态学算子结合面向对象多尺度分割方法,简化原始影像复杂的背景信息,快速提取大范围单木树冠信息。以甘肃省张掖市大野口林区机载激光雷达系统携带的CCD影像为数据源,提取实验区单木树冠,并从空间和形状上验证效果。实验结果表明:在高分影像的灰度梯度图像上进行面向对象分割提取单木冠幅,单木株数精度为83.19%,形状精度达到88.62%,优于传统林业调查精度,且冠幅获取速度快,效率高,并可以较为精确地提取树冠边界。
[Abstract]:Tree crown is an important part of trees. Tree crown extraction based on remote sensing image is of great significance for forest resources investigation and monitoring, but it is difficult to obtain the shape and boundary of tree crown accurately. High-resolution images have abundant texture and spectral information. Based on high-resolution images single tree crown delineation technology provides a rapid and effective approach for forest resource survey. However, due to the complicated information of the high-score image, the object-oriented segmentation method has a large amount of computation, and it needs to set the spectral or texture threshold manually to achieve the single tree segmentation, which results in the decrease of the work efficiency and the poor robustness. Image enhancement can selectively highlight or suppress some features of the original image by changing the structure of the original image. Effective image enhancement is beneficial to improve the accuracy of single tree crown segmentation. Therefore, this paper proposes a tree crown extraction method based on image grayscale gradient image segmentation. By comparing the traditional Robles, Laplacian operator with the improved mathematical morphology operator, this paper proposes a tree crown extraction method based on image grayscale gradient image segmentation. The improved mathematical morphology operator is determined by the combination of visual interpretation and gray-scale histogram. Then, the improved mathematical morphology operator combined with the object-oriented multi-scale segmentation method is used to simplify the complex background information of the original image and quickly extract the large-scale single tree crown information. The CCD image carried by airborne lidar system in Dayakou forest area of Zhangye City Gansu Province was used as data source to extract the single tree crown in the experimental area and verify the effect from space and shape. The experimental results show that the object-oriented segmentation and extraction of crown width from grayscale gradient images of high-score images shows that the accuracy of individual tree number and shape is 83.19% and 88.62% respectively, which is better than that of traditional forestry survey. Moreover, the crown width can be obtained quickly and efficiently, and the crown boundary can be extracted accurately.
【作者单位】：北京林业大学林学院北京林业大学精准林业北京市重点实验室省部共建森林培育与保护重点实验室;
【基金】：国家重大科学仪器设备开发专项(2013YQ12034304)
【分类号】：S771;TP391.41

【参考文献】