基于变量优选的机载激光雷达对林分平均高的反演  

Inversion of average forest stand height based on variable selection by airborne laser radar

作　　者：朱泊东 金京 罗洪斌 龙飞 李春干[2] 岳彩荣 ZHU Bo-dong;JIN Jing;LUO Hong-bin;LONG Fei;LI Chun-gan;YUE Cai-rong(College of Forestry,Southwest Forestry University,Kunming 650224,China;College of Forestry,Guangxi University,Nanning 530004,China)

机构地区：[1]西南林业大学林学院,云南昆明650224 [2]广西大学林学院,广西南宁530004

出　　处：《江苏农业学报》2022年第3期706-713,共8页Jiangsu Journal of Agricultural Sciences

基　　金：国家自然科学基金项目(42061072);云南省科技厅重大科技专项(202002AA00007-015);云南省教育厅项目(2018JS330)。

摘　　要：森林高度是反映森林数量和质量的重要指标,是森林经营管理的重要基础数据,准确获取森林高度信息一直是林业遥感研究的目标。本研究以广西高峰林场的105块地面实测样地数据和机载激光雷达(Light detection and ranging,LiDAR)数据为基础,从点云数据中提取35个特征变量,分别采用支持向量机-递归特征消除法(SVM-RFE)、轻量级梯度提升机(LightGBM)和主成分分析(PCA)法进行特征筛选,并结合参数模型(LR)和非参数模型(RFR、KNN)对林分平均高进行反演。研究结果表明,不同特征选择方法和估测模型的组合精度差异较大。其中,利用LightGBM进行特征筛选结合KNN回归反演效果最佳,建模的R^(2)和RMSE分别为0.83和1.64 m,验证的R^(2)和RMSE分别为0.81和1.56 m。此外,在SVM-RFE、LightGBM和PCA这3种特征筛选方法中LightGBM的效果最好,无论在RFR模型还是在KNN模型中均能得到较高的R^(2),优于SVM-RFE和PCA。Forest height is an important indicator of forest quantity and quality,and it is also an important essential parameter for forest management.Obtaining the height information of the forest accurately has always been the target of remote sensing study of the forestry.In this study,35 characteristic variables were extracted from the point cloud data,based on the measured data from 105 sample ground plots and airborne light detection and ranging(LiDAR)data of Guangxi State-owned Gaofeng Forest Farm.Support vector machine-recursive feature elimination(SVM-RFE),light gradient boosting machine(LightGBM)and principal components analysis(PCA)were used to screen the characteristics,respectively.The average height of forest were inverted by combining parametric model(LR)and nonparametric models(RFR,KNN).The results showed that,there were large differences of accuracy between combinations of different feature selection methods and estimation models.Among them,the combination of LightGBM feature selection with KNN regression inversion showed the best effect.The R^(2) and root mean square error(RMSE)of modeling were 0.83 and 1.64 m,respectively,and the R^(2) and RMSE of verification were 0.81 and 1.56 m,respectively.In addition,among the three feature selection methods of SVM-RFE,LightGBM and PCA,LightGBM showed the best effect,which was high in R^(2) both in the RFR model and in the KNN model,and was better than SVM-RFE and PCA.

关 键 词：机载激光雷达 林分平均高 特征优选 LightGBM 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]