Abstract: Six image segmentation algorithms based on the pixel clustering and watershed thought are chosen for segmenting the orthographic images of Chinese pine and sea-buckthorn captured by unmanned aerial vehicle. As evaluation criteria of algorithms, misclassification error, relative ultimate measurement accuracy and running time are implemented to compare segmentation performance of each algorithm objectively and quantitatively. Experimental results demonstrate that the accuracy of each algorithm is closely related to the image capture height and noise. The conclusion can play a guidance role in selecting algorithm for stricken forest orthographic images segmentation.

Key words: forest diseases and insect pests, unmanned aerial vehicle orthographic image, image segmentation, algorithms based on pixel clustering, watershed algorithm, algorithms&rsquo, performance evaluation

摘要： 以错分率、相对最终测量精度以及运行时间为评价标准，利用无人机采集的油松及沙棘正射图像为测试图像，对6种基于像素聚类及分水岭的图像分割算法的性能进行了定性分析及定量比较。实验结果表明，受灾林区图像的分割算法的性能与图像拍摄高度、噪声等因素密切相关。最后，给出了受灾林区无人机正射图像分割算法应用的指导性建议。

关键词: 森林病虫害监测, 无人机正射图像, 图像分割, 基于像素的聚类, 分水岭算法, 性能评价