林木冠层图像的分割方法研究

Segmentation method research for forest canopy image

现代林业研究中,有效的提取并分析图像数据中树木信息,对树木深层信息挖掘起着很重要的作用。实际上,从复杂的林业图像中准确的分割出目标植物是很多后续图像研究的关键问题之一。大津法是目前使用最为广泛的图像分割算法之一,但是由于树木图像的自然性,该方法对冠层图像临界处分割往往存在较大误差。为解决该问题,本文结合类间方差及类内聚度这两个度量值,改进大津法阈值选取目标函数,并且以银杏冠层图像为例进行分割,结果表明:(1)对于郁闭度较小的冠层图像,两种分割方法得到的分割效果较为相近;(2)对于郁闭度较大的冠层图像,本文改进方法较传统大津法可得到更好的分割效果。(3)改进法分割冠层图像是可行的。

How to extract and analyze the tree information from the image data is very important for deeply excavating the deeper-level information of trees and forests in modern forestry researches.In fact,it is one of the crucial issues in the subsequent image research that accurately segment the target plant from complicated forest image.Among all the image segmentation methods,Otsu method is one of the most widely used threshold selection methods of image segmentation because of its simple calculation and great adaptability.However,due to the naturality of the tree images,it is not able to obtain good segmentation results when the canopy images were segmented by using Otsu method.Considering its poor segmentation results of the canopy images,an improved threshold selection algorithm based on traditional Otsu method was put forward.Both between-class variance and within-class variance were considered that impact on the image segmentation result,the selection of objective function of Otsu method was improved and the image segmentation was conducted by taking the Ginkgo biloba canopy images as the example.The results show（1） for the images of lower canopy density,the results obtained by traditional Otsu method and improved Otsu method were similar each other;（2） for the images of higher canopy density,better segmentation results can be obtained by using the improved method rather than traditional Otsu method;（3） according to the results,it is feasible to segment the canopy images by using the improved Otsu segmentation method.