CAJ | 학술논문

以湖南省攸县黄丰桥林场Worldview-2影像和地面样地调查数据为基础，采用Mean shift算法对影像进行多尺度分割，提取杉木人工林林木冠幅信息，共提取有效林木冠幅227个，并对提取的冠幅边界信息进行平滑处理。分析调查数据中实测冠幅与影像提取冠幅之间的相关性，结合实测胸径、树高与冠幅的关系，应用曲线估计、非线性联立方程组以及基于哑变量的非线性联立方程组分别建立树高和胸径的最优估算模型，并进行了精度评价。结果表明：将树高与胸径作为哑变量，并进行数量化分级建立的影像冠幅与胸径、树高的非线性误差变量联立方程组模型的拟合效果要优于其他2种方法，树高和胸径模型决定系数R2H和R2D分别为0.899和0.913。模型的适用性检验表明，模型的变动系数、平均百分标准误差均在10%以内，具有较强的稳健性。
이호남성유현황봉교림장Worldview-2영상화지면양지조사수거위기출，채용Mean shift산법대영상진행다척도분할，제취삼목인공림림목관폭신식，공제취유효림목관폭227개，병대제취적관폭변계신식진행평활처리。분석조사수거중실측관폭여영상제취관폭지간적상관성，결합실측흉경、수고여관폭적관계，응용곡선고계、비선성련립방정조이급기우아변량적비선성련립방정조분별건립수고화흉경적최우고산모형，병진행료정도평개。결과표명：장수고여흉경작위아변량，병진행수양화분급건립적영상관폭여흉경、수고적비선성오차변량련립방정조모형적의합효과요우우기타2충방법，수고화흉경모형결정계수R2H화R2D분별위0.899화0.913。모형적괄용성검험표명，모형적변동계수、평균백분표준오차균재10%이내，구유교강적은건성。
A state-owned Huang-Feng-Qiao Forest Farm in Youxian County, Hunan Province was chosen as the study area. Based on Worldview-2 remote sensing data and ground sample survey data of the forest farm, and by adopting Mean shift algorithm, the canopy information of plantation forest of Chinese ifr in the farm were extracted with multi-scale segmentation method. Totally, 227 canopy breadth information of effective tested tree were extracted and the tree crown width and crown boundary information extracted were smoothed. The correlation between measured crown width and the crown extracted from the images was studied. By taking into account the relationship between diameter at breast height, tree height and crown width measured, applying curve estimation, nonlinear equations and dummy variable non-linear simultaneous equations, the optimal estimation models of tree height and diameter at breast height were respectively established. Furthermore, the precisions of the model estimation were evaluated. The results show that with the DDBH and HTH as dummy variables respectively, they were graded quantitatively, thus creating a non-linear simultaneous equations model. The iftting effect were much better than those of the ifrst two methods, and the determination coefifcients of tree height and diameter at breast height model (R2H, R2D) were 0.899 and 0.913 respectively. The adaptability testing of the models showed that Standard Error of Estimate, Coefifcient of Variation and Mean Percent Standard Error all were less than 10%, with strong robustness.