基于无人机影像的天山云杉林更新类别划分及提取

Classification and Extraction of Updating Categories of Picea schrenkiana var.tianschanica Based on UAV Images

森林伐后更新对森林资源的有效利用密不可分,对森林的经营管理和动态监测具有重要意义。为了对不同皆伐迹地林分进行主要特征选取并实现高精度提取,本研究通过分析天山云杉皆伐迹地更新群落特征,并利用R软件进行聚类分析,将其划分为4个类别(Ⅰ、Ⅱ、Ⅲ、Ⅳ),其林分平均年龄为33、28、25 a和20 a;平均胸径分别为25.62、17.95、13.03 cm和10.05 cm;平均树高为12.25、10.02、8.92 m和7.53 m。基于eCognition Developer 9.0软件,对研究区进行多层次、多尺度分割,以无人机影像的纹理、光谱等特征为辅助信息进行SEaTH算法结合面向对象分类。结果表明,研究区皆伐迹地林分生长状况主要为类别Ⅰ状态,类别Ⅱ、Ⅲ和Ⅳ分布面积相对较少。SEaTH算法结合隶属度分类方法,不仅准确地识别出皆伐迹地空间分布情况,还精确地提取出不同类别的天山云杉林,总体精度达到80.22%,Kappa系数为0.73,平均面积吻合度为81.34%。该方法取得了较好的分类效果,可精确地对伐后更新林分进行分级提取,为森林经营管理提供一定的参考依据。

The forest regeneration after clearing cutting is closely related to the effective utilization of forest resources,which is of great significance to forest management and dynamic monitoring.In order to select and realize the high-precision extraction of the main characteristics of different cutting blanks,the characteristics of the community of the cutting blanks of Picea schrenkiana var.tianschanica were analyzed.After cluster analysis by R software,the cutting blanks were divided into four categories:Ⅰ,Ⅱ,Ⅲ,andⅣwith the mean forest ages of 33,28,25,and 20 a;mean diameters at breast height of 25.62,17.95,13.03,and 10.05 cm;mean heights of 12.25,8.925,10.02,and 7.53 m,respectively.Based on the eCognition Developer 9.0 software,the multi-level and multi-scale segmentations of the research area were carried out,and the SEaTH algorithm was combined with the object-oriented classification based on the characteristics of the texture,the spectrum and the like of the unmanned aerial vehicle image.The results showed that the growth of the forest stand in the cutting blanks was mainly in the categoryⅠ,and the distribution areas of the categoryⅡ,ⅢandⅣwere relatively small.The SEaTH algorithm,combined with the membership classification method,could not only accurately identify the spatial distribution of the whole area,but also accurately extract the different types of the P.schrenkiana var.tianschanica forest with the overall precision,the Kappa coefficient,and the average area matching degree of 80.22%,0.73,and 81.34%,respectively.In conclusion,the method has a good classification effect,and can be used for grading and extracting the stand after cutting,and provides a reference for forest management.