[1]朱泊东金京,罗洪斌,龙飞,等.基于变量优选的机载激光雷达对林分平均高的反演[J].江苏农业学报,2022,38(03):706-713.[doi:doi:10.3969/j.issn.1000-4440.2022.03.016]
　ZHU Bo-dong,JIN Jing,LUO Hong-bin,et al.Inversion of average forest stand height based on variable selection by airborne laser radar[J].,2022,38(03):706-713.[doi:doi:10.3969/j.issn.1000-4440.2022.03.016]
点击复制

基于变量优选的机载激光雷达对林分平均高的反演()
分享到：

江苏农业学报[ISSN:1006-6977/CN:61-1281/TN]

卷:
38
期数:
2022年03期
页码:
706-713
栏目:
农业信息工程
出版日期:
2022-06-30

文章信息/Info

Title:
Inversion of average forest stand height based on variable selection by airborne laser radar
作者:
朱泊东1金京1; 罗洪斌1; 龙飞1; 李春干2; 岳彩荣1
(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)
Author(s):
ZHU Bo-dong1; JIN Jing1; LUO Hong-bin1; LONG Fei1; LI Chun-gan2; YUE Cai-rong1
(1.College of Forestry, Southwest Forestry University, Kunming 650224, China;2.College of Forestry, Guangxi University, Nanning 530004, China)
关键词:
机载激光雷达; 林分平均高; 特征优选; LightGBM
Keywords:
airborne laser radar; average forest stand height; feature selection; light gradient boosting machine (LightGBM)
分类号:
TP79
DOI:
doi:10.3969/j.issn.1000-4440.2022.03.016
文献标志码:
A
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R2和RMSE分别为0.83和1.64 m，验证的R2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R2，优于SVM-RFE和PCA。
Abstract:
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 R2 and root mean square error (RMSE) of modeling were 0.83 and 1.64 m, respectively, and the R2 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 R2 both in the RFR model and in the KNN model, and was better than SVM-RFE and PCA.

参考文献/References:

[1]董立新.林分平均高度卫星遥感新进展[J].遥感技术与应用,2016,31(5):833-845.
[2]ZHAO F, GUO Q, KELLY M. Allometric equation choice impacts lidar-based forest biomass estimates: a case study from the Sierra National Forest, CA[J]. Agricultural and Forest Meteorology, 2012, 165: 64-72.
[3]刘亚男.基于多源遥感数据的森林地上生物量及净初级生产力估算研究[J].测绘学报,2020,49(12):1641.
[4]GREGOIRE T G , NSSET E , MCROBERTS R E , et al. Statistical rigor in LiDAR-assisted estimation of aboveground forest biomass[J]. Remote Sensing of Environment, 2016, 173:98-108.
[5]PEDERSEN R , BOLLANDSS O M, GOBAKKEN T, et al. Deriving individual tree competition indices from airborne laser scanning[J]. Forest Ecology and Management, 2012, 280: 150-165.
[6]刘鲁霞,庞勇.机载激光雷达和地基激光雷达林业应用现状[J].世界林业研究,2014,27(1):49-56.
[7]穆喜云, 张秋良, 刘清旺, 等. 基于机载LiDAR数据的林分平均高及郁闭度反演[J]. 东北林业大学学报, 2015, 43(9): 84-89.
[8]彭涛,邢艳秋,尤号田，等．LiDAR采样形状及采样尺度对林分均高估测的影响[J].中南林业科技大学学报,2018,38(4)：27-32．
[9]曾伟生,孙乡楠,王六如,等. 基于机载激光雷达数据估计林分蓄积量及平均高和断面积[J].林业资源管理,2020(2):79-86.
[10]焦义涛,邢艳秋,霍达,等. 基于机载LiDAR点云估测林分的平均树高[J]. 西北林学院学报, 2015, 30(3): 170-174.
[11]赵勋,岳彩荣,李春干,等. 基于机载 LiDAR 数据估测林分平均高[J]. 林业科学研究, 2020, 33(4): 59-66.
[12]周蓉,赵天忠,吴发云.依据BP神经网络的机载LiDAR数据估算林分平均高[J].东北林业大学学报,2021,49(9):60-66.
[13]沈剑波,雷相东,李玉堂,等.基于BP神经网络的长白落叶松人工林林分平均高预测[J].南京林业大学学报(自然科学版),2018,42(2):147-154.
[14]GRACE Y，HE W Q，CARROLL R J. Feature screening with large scale and high dimensional survival data.[J]. Biometrics, 2021,77(2):1-14.
[15]曹林,代劲松,徐建新,等. 基于机载小光斑 LiDAR 技术的亚热带森林参数信息优化提取[J]. 北京林业大学学报, 2014, 36(5): 13-21.
[16]PAOLO M, VIBRANS A C, MCROBERTS R E, et al. Methods for variable selection in LiDAR-assisted forest inventories[J]. Forestry,2017, 90: 112-124.
[17]郝红科. 基于机载激光雷达的森林参数反演研究[D]. 杨凌：西北农林科技大学, 2019.
[18]莫莉婕.国有林场森林资源可持续发展对策研究——以高峰林场为例[J].企业科技与发展,2021(4):222-224.
[19]GUYON I, WESTON J, BARNHILL S, et al. Gene selection for cancer classification using support vector machines[J]. Machine Learning, 2002, 46(1): 389-422.
[20]KE G L, MENG Q, FINLEY T, et al. LightGBM: a highly efficient gradient boosting decision tree[J]. Advances in Neural Information Processing Systems, 2017, 30: 3146-3154.
[21]BREIMAN L. Random forests[J].Machine Learning，2001,45 (1) :5-32．
[22]ACAR L. Some examples for the decentralized receding horizon control[C]. Tucson: IEEE, 1992: 1356-1359.
[23]郝玲,张佩,史逸民,等. 基于机器学习的江苏省冬小麦气象产量客观区划及歉年预测[J].江苏农业科学,2021,49(12):162-168.
[24]白婷,丁建丽,王敬哲. 基于机器学习算法的土壤有机质质量比估算[J].排灌机械工程学报,2020,38(8):829-834.
[25]赵献立,王志明. 机器学习算法在农业机器视觉系统中的应用[J].江苏农业科学,2020,48(12):226-231.

备注/Memo

备注/Memo:
收稿日期：2021-10-29基金项目：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)作者简介：朱泊东(1997-),男,云南曲靖人,硕士研究生,主要从事林业遥感研究。（E-mail）zhubodong@swfu.edu.cn通讯作者：岳彩荣,(E-mail)cryue@163.com
更新日期/Last Update: 2022-07-07