基于无人机可见光影像的树种和树冠信息提取——以晋西黄土区蔡家川流域为例

为探究无人机遥感技术在黄土高原森林资源调查中的适用性,以晋西黄土区蔡家川流域为研究区,以无人机可见光影像为遥感数据源,基于面向对象最邻近分类法,识别并提取研究流域的树种和树冠信息,并与样方调查数据进行对比分析,评估无人机影像提取植被信息的精度及其适用性。结果表明:面向对象最邻近分类法对于郁闭度较低的林分和经济果木林的树种提取效果极好,但复杂植被类型会导致提取精度下降。在农地子流域和人工林子流域上,树种提取的分类混淆矩阵Kappa系数分别为0.898和0.728。面向对象最邻近分类法对人工林和经济果木林的树冠提取精度较高,与实测数据线性回归的决定系数(R²)在0.7以上,但对次生林的树冠提取效果相对较差,R²仅有0.422 3。将该方法拓展应用至流域尺度,识别结果显示,蔡家川流域内人工林子流域主要为刺槐、油松和侧柏混交林,经济作物主要为苹果,油松的林分密度为1 744株·hm^-2,平均冠幅为2.24 m,苹果的林分密度为382株·hm^-2,平均冠幅为4.26 m;农地子流域有苹果树912株,林分密度为439株·hm^-2,平均冠幅为3.84 m。结果表明,基于无人机遥感影像,利用面向对象最邻近分类法可以高效、准确地提取林木株数、郁闭度和平均冠幅,从而有效提高黄土区植被调查的效率。

Determining tree species and crown width from unmanned aerial vehicle imagery in hilly loess region of west Shanxi,China: a case study from Caijiachuan watershed

In the present study, Caijiachuan Watershed in the hilly loess region of west Shanxi was selected as the study area to examine the feasibility of unmanned aerial vehicle (UAV) remote sensing in extracting forest species and stand characteristics. The information of forest species and crown width were identified and extracted from visible images of UAV using the object-based nearest neighbor classification method, and the accuracy was evaluated by comparison with field investigation. The presented method was also used in the watershed level to investigate forest species and canopy characteristics in the farmland sub-watershed and plantation sub-watershed. The results showed that the Kappa coefficient of the confusion matrix for forest species classification was 0.898 and 0.728 in the farmland sub-watershed and in the plantation sub-watershed, respectively, suggesting higher accuracy under the condition of low canopy density than that under multiple vegetation types with high canopy density. According to the determination coefficient between the extracted and measured crown width, the accuracy of crown width extraction of planted forests and economic forest (R²>0.7) was higher than that of secondary forests (R²=0.422 3). Based on the UAV remote sensing and the presented method, it was found that the mixed forest of Robinia pseudoacacia L., Pinus tabulaeformis Carr. and Platycladus orientalis(L.) Franco was dominant, along with the economic forest, Malus domestica Borkh. in the plantation sub-watershed, as the stand density of Pinus tabulaeform was 1 744 hm ^-2, with the average crown width of 2.24 m, ad the stand density of Malus domestica was 382 hm ^-2 with the average crown width of 4.26 m. There were 912 Malus domestica, with a stand density of 439 hm ^-2 and an average crown width of 3.84 m in the farmland sub-watershed. These results indicated that the proposed method coupled with UAV remote sensing would improve the efficiency and increase the accuracy of forest resource surveys with the capability in forest species classification, tree number counting and crown width estimation.