欢迎访问林业科学,今天是

林业科学 ›› 2019, Vol. 55 ›› Issue (4): 187-195.doi: 10.11707/j.1001-7488.20190420

• 研究简报 • 上一篇    下一篇

基于线性谱聚类的林地图像中枯死树监测

宋以宁1, 刘文萍1, 骆有庆2, 宗世祥2   

  1. 1. 北京林业大学信息学院 北京 100083;
    2. 北京林业大学林学院 北京 100083
  • 收稿日期:2018-03-20 修回日期:2018-07-18 出版日期:2019-04-25 发布日期:2019-04-30
  • 基金资助:
    “十三五”国家重点研发计划“林业有害生物检测、监测与预警关键技术”(2018YFD0600201);北京市科技计划项目“影响北京生态安全的重大钻蛀性害虫防控技术研究与示范”(Z171100001417005);中央高校基本科研业务费专项资金(2015ZCQ-XX)。

Monitoring of Dead Trees in Forest Images Based on Linear Spectral Clustering

Song Yining1, Liu Wenping1, Luo Youqing2, Zong Shixiang2   

  1. 1. College of Information, Beijing Forestry University Beijing 100083;
    2. College of Forestry, Beijing Forestry University Beijing 100083
  • Received:2018-03-20 Revised:2018-07-18 Online:2019-04-25 Published:2019-04-30

摘要: [目的]将基于线性谱聚类超像素的方法应用在森林病虫害防治领域,可智能监测无人机森林虫害图像中的枯死树,为森林有害生物的监测工作提供技术支撑。[方法]分别以湖北省受松材线虫与辽宁省受红脂大小蠹侵害的松林无人机图像为试验数据,首先使用线性谱聚类超像素分割算法将图像划分为多个超像素;然后基于枯死树木的颜色特征,初步提取可能为枯死树的超像素区域;最后基于枯死树木与其他干扰地物具有不同的纹理特征,计算超像素的区域密度和缝隙量,利用支持向量机对初步提取的超像素进行分类,从而检测出图像中的枯死树。[结果]基于线性谱聚类超像素和支持向量机的枯死树监测方法可有效排除与枯死树木颜色相近的其他干扰地物,较准确地提取出枯死树木。使用该方法与基于植被颜色指数的阈值分割方法、基于简单线性迭代聚类超像素和随机森林的方法,对35幅受灾松林无人机图像进行试验,并选用交并比、虚警率和漏检率3个评价指标对3种方法进行定量对比分析。结果表明,基于线性谱聚类超像素的方法监测出的枯死树区域最精确,其监测结果与人工检测结果的交并比均值大于58%,且虚警率和漏检率均优于另外2种方法。[结论]基于线性谱聚类超像素的枯死树监测方法能实现松林中枯死树的快速、准确检测及定位。

关键词: 无人机图像分析, 森林病虫害, 枯死树监测, 纹理特征提取, 超像素, 线性谱聚类

Abstract: [Objective] In this paper, the method based on linear spectral clustering (LSC) superpixel was applied in the field of forest pest control, which was able to intelligently monitor dead trees in forest pest images taken from the unmanned aerial vehicle (UAV), and provide technological support for intelligently monitoring of forest pests.[Method]The UAV images of pine forests infected by Bursaphelenchus xylophilus and Dendroctonus valens respectively from Hubei and Liaoning provinces were chosen as the experiment data. Firstly, the linear spectral clustering superpixel algorithm was used to divide the image into many compact and uniform superpixels. Then, on a basis of the different color characteristics of dead trees and healthy trees, the superpixels which might be dead trees were initially extracted. Next, based on the different texture features of dead trees and other red disturbances, the area density and the lacunarity of the initially extracted superpixels were calculated. Finally, the support vector machine based on texture features was used to classify the initially extracted superpixels to detect dead trees in the image.[Result]The method based on LSC superpixel was able to exclude other interference objects that were similar in color to dead trees, and accurately extracted dead trees. The 35 UAV images of the pest-infected pine forest were used for comparing quantitatively this method with the other two methods. One is threshold segmentation method based on vegetation color index, and the other is simple linear iterative clustering (SLIC) superpixel and random forest method. Furthermore, the three evaluation indexes:intersection over union, the false alarm rate and the misdetection rate were used to quantitatively compare and analyze the three methods. The experimental results showed that the algorithm based on LSC superpixel and SVM was the most accurate to detect dead trees. The mean of intersection over union between this method's result and manual detection result was more than 58% and the false alarm rate and the misdetection rate were better than the other two algorithms.[Conclusion] Our results showed that the dead tree monitoring method based on LSC superpixel was able to detect and locate dead trees quickly and precisely in the UAV pine forest images and effectively protect forest resources.

Key words: analysis of unmanned aerial vehicle image, forest pest, dead tree monitoring, texture feature extraction, superpixel, linear spectral clustering

中图分类号: