基于级联循环网络的林木生长参数预测

黄成威; 齐磊; 多杰才仁; 张怀清; 薛联凤; 云挺

doi:10.12171/j.1000-1522.20230027

基于级联循环网络的林木生长参数预测

doi: 10.12171/j.1000-1522.20230027

黄成威^1,,
齐磊²,
多杰才仁³,
张怀清⁴,
薛联凤¹,
云挺^{1, 3, ,}

1.
南京林业大学信息科学技术学院、人工智能学院，江苏南京 210037
2.
北京林业大学生态与自然保护学院，北京 100083
3.
南京林业大学林草学院、水土保持学院，江苏南京 210037
4.
中国林业科学研究院，北京 100091

基金项目: 国家自然科学基金项目（31770591、32071681），江苏省自然科学基金面上项目（BK20221337），江苏省农业自主创新项目（CX(22)3048），自然资源部国土卫星遥感应用重点实验室开放基金项目（KLSMNR-G202208）

详细信息

作者简介:
黄成威。主要研究方向：林业人工智能。 Email：745164733@qq.com　地址：210037 江苏省南京市玄武区龙蟠路159号南京林业大学信息科学技术学院

责任作者:
云挺，博士，教授。主要研究方向：林业人工智能、林业元宇宙与数字孪生。 Email：njyunting@qq.com　地址：同上

中图分类号: S718.5；TP183
计量
- 文章访问数: 113
- HTML全文浏览量: 38
- PDF下载量: 50
- 被引次数: 0
出版历程
- 收稿日期: 2023-02-03
- 修回日期: 2023-06-06
- 录用日期: 2023-07-05
- 网络出版日期: 2023-07-07
- 刊出日期: 2023-08-25

Prediction of tree growth parameters based on cascaded recurrent network

1.
School of Information Science and Technology, School of Artificial Intelligence, Nanjing Forestry University, Nanjing 210037, Jiangsu, China
2.
School of Ecology and Nature Conservation, Beijing Forestry University, Beijing 100083, China
3.
School of Forestry and Grass, School of Water and Soil Conservation, Nanjing Forestry University, Nanjing 210037, Jiangsu, China
4.
Chinese Academy of Forestry, Beijing 100091, China

摘要

摘要: 目的树木的生长参数在林木性状评估、森林碳汇计量和优化森林经营策略等方面具有重要的意义。从激光雷达数据中精准提取林木生长参数并对树木未来生长参数进行预测，以期为林业数字化发展提供技术支持。方法本研究提出了一种基于人工智能的林木生长参数提取与预测方法，该方法以南京林业大学中的樱花、银杏、鹅掌楸、水杉4个树种为实验对象。首先，采用机载激光雷达获取4个树种样地的点云数据，并通过单株分割算法提取单棵树木点云。其次，基于圆拟合及高斯滤波的方法自动的从2016、2018、2020年的单棵树点云中提取胸径、树高、冠宽等生长参数，并辅以样地调查数据（2015、2017、2019、2021、2022年），构建不同树种的单棵树生长参数时间序列作为深度学习网络的训练样本集。最后，构造由两层门控循环单元（GRU）的林木生长参数预测深度学习网络，并引入注意力模块以弥补传统循环神经网络捕获长期依赖关系的不足。该网络以2015—2021年单株树木生长参数时间序列作为输入，依托训练数据及随机梯度下降算法使网络参数逼近真实树木生长情况，并用以预测2022年单棵树木的生长参数。结果深度学习网络在树高预测上表现最好，决定系数R²均不低于0.83，均方根误差（E_RMS）均小于0.50 m，在4种树中银杏树的预测结果最优（R² = 0.95，E_RMS = 0.31 m）。在胸径、冠宽等参数的预测上，深度预测网络仍有着良好的表现，R²均不低于0.81，胸径E_RMS小于2.50 cm，冠宽E_RMS小于0.32 m。在与线性回归和LSTM网络等林木参数预测方法的比较中，本模型预测效果良好（R² ≥ 0.86），误差较小。结论级联循环神经网络可有效地预测未来树木的生长情况，提高林木生长参数的预测精度，同时GRU和注意力机制的引入在林木参数的时序预测中有一定的鲁棒性，为森林的智能管理与可视化分析提供了新的思路。
- 激光点云(LiDAR) /
- 深度学习 /
- 智慧林业 /
- 林木生长参数预测 /
- 级联循环网络
Abstract: Objective Accurate acquisition of tree growth parameters is a necessary prerequisite for obtaining forest information and phenotypic characteristics and is of great significance in forest character assessment, forest carbon sink measurement, and optimization of forest management strategies. Nowadays, accurate extraction of tree growth parameters from light detection and ranging data and prediction of tree growth parameters in the future will provide technical support for the digital development of forestry. Method In this study, an artificial intelligence based approach to tree growth parameter extraction and prediction was proposed. Four tree species in Nanjing Forestry University, i.e., sakura, ginkgo, liriodendron, and Chinese fir, were selected as experimental objects. First, airborne laser scanning was used to obtain the point cloud data of four tree species sample plots, and the individual tree segmentation algorithm was used to obtain a single tree point cloud. Second, growth parameters such as DBH height, tree height, and crown width were automatically extracted from the single tree point cloud using the circle fitting and Gaussian filter algorithms, and supplemented with artificial in-situ measurement data (measured in 2015, 2017, 2019, 2021, and 2022). The time series of single tree growth parameters of different tree species were constructed as the training sample set for the deep learning network. Finally, a deep learning network for predicting tree growth parameters was constructed with a two-layer Gated Recurrent Unit (GRU) network, and an attention module was introduced to compensate for the shortcomings of traditional recurrent neural networks in capturing long-term dependencies. The network took the time series of individual tree growth parameters from 2015 to 2021 as the input, relied on training data and stochastic gradient descent algorithm to approximate the network parameters to the real tree growth situation, and was used to predict the growth parameters of individual trees in 2022. Result The network performed best in the prediction of tree height, with the R² not less than 0.83 and root mean square error (E_RMS) less than 0.50 m. The prediction results of the ginkgo tree were the best among the four species (R² = 0.95, E_RMS = 0.31 m). The depth prediction network still had an acceptable performance in the prediction of parameters such as DBH height and crown width, R² was not less than 0.81, DBH height E_RMS was less than 2.50 cm, crown width E_RMS was less than 0.32 m. The model in this paper predicted well (R² ≥ 0.86) with less error compared with other tree parameter prediction methods such as linear regression and LSTM networks. Conclusion The cascaded recurrent neural network can effectively predict the future tree growth and improve the prediction accuracy of tree growth parameters, while the introduction of GRU and attention model has certain robustness in the time-series prediction of growth parameters, which provides a new idea for the intelligent management and visualization analysis of forests.
- light detection and ranging (LiDAR) /
- deep learning /
- intelligent forestry /
- forest growth parameter prediction /
- cascaded recurrent network

HTML全文

图 1 本研究流程图

Figure 1. Flowchart of this study

下载: 全尺寸图片幻灯片

图 2 南京林业大学校园研究区域示意图

Figure 2. Schematic diagram of the study area of Nanjing Forestry University campus

下载: 全尺寸图片幻灯片

图 3 本研究构造的用于树木生长参数预测的深度学习网络

Figure 3. Conceived deep learning network to predict growth parameters of the study trees

下载: 全尺寸图片幻灯片

图 4 部分样地树顶检测和树冠分割示意图

a. 水杉；b. 樱花；c. 鹅掌楸；d. 银杏样地。字母后地数字1、2、3分别对应于激光点云数据的3个阶段（即2016年、2018年和2020年研究区域的点云数据）。4种树中樱花的高度大致为5 ~ 8 m，故在统一的相对高度色条下高程颜色显示及变化不明显。a, Chinese fir; b, sakura; c, liriodendron; d, ginkgo. Numbers after alphabet of 1, 2 and 3 are correspond to the three phases of the point cloud data (i.e., the data from study sites in 2016, 2018, and 2020). It should be noted that the height of sakura is approximately 5–8 m, so the elevation color changes are not very apparent under the unified elevation color bar.

Figure 4. Schematic diagram of treetop detection and tree crown segmentation in some study sites

下载: 全尺寸图片幻灯片

图 5 单株分离和胸径拟合示意图

每个树种展示了18棵树的完整点云和4棵树的胸径拟合结果，即：a、b、A为水杉，c、d、B为樱花，e、f、C为银杏，g、h、D为鹅掌楸。同一棵树在2016、2018和2020年扫描的点云数据在每个三维坐标轴区域显示。其中，粉色、绿色和蓝色点云分别代表2016、2018和2020年的点云数据。黑色代表地面点云。每个三维坐标轴的右上角由上至下分别标注了2020、2018和2016年从激光点云中测得的树木高度，单位为m。Each tree species shows 18 trees with complete point cloud and the DBH fitting results of four trees, namely a, b, A of Chinese fir, c, d, B of sakura, e, f, C of ginkgo, g, h, D of liriodendron. Point cloud data collected by the same tree in 2016, 2018 and 2020 are displayed in each axis region. Among them, hot pink, green and blue represent point cloud data in 2016, 2018 and 2020, respectively. Black represents ground point clouds. The height of trees measured from point cloud in 2020, 2018 and 2016 is marked in m from top to bottom in the upper right corner of each coordinate axis.

Figure 5. Schematic diagram of individual plant isolation and DBH fitting

下载: 全尺寸图片幻灯片

图 6 本文构建的用于预测生长参数的深度学习网络在不同批量大小下的损失函数曲线

Figure 6. Curves of training loss value of deep learning network for growth parameter prediction under different batch sizes

下载: 全尺寸图片幻灯片

图 7 测试数据中3种树木生长参数的预测结果与实地测量数据的对比

Figure 7. Scatter plots illustrating the comparison results of the tree growth parameters obtained by field measurements versus our method for the three different parameter types

下载: 全尺寸图片幻灯片

表 1 从激光点云和样地测量中获取的4种树的在不同年份的生长参数

Table 1. Summary of tree growth parameters obtained from point cloud data and field measurements for four tree species in different years

数据来源 Data sources	生长参数 Growth parameter	水杉 Chinese fir				樱花 Sakura				银杏 Ginkgo				鹅掌楸 Liriodendron
数据来源 Data sources	生长参数 Growth parameter	最大值 Max. value	最小值 Min. value	平均值 Mean	SD	最大值 Max. value	最小值 Min. value	平均值 Mean	SD	最大值 Max. value	最小值 Min. value	平均值 Mean	SD	最大值 Max. value	最小值 Min. value	平均值 Mean	SD
样地测量（2015） Sample plot measurement (2015)	W_Cs-n/m	7.95	3.03	5.42	2.61	6.31	4.17	5.48	0.60	6.97	4.52	5.18	1.17	7.94	5.15	6.21	0.99
	W_Ce-w/m	7.13	2.51	4.93	2.77	7.83	4.24	6.53	1.19	4.73	2.83	3.72	0.79	6.53	4.17	5.34	0.80
	H/m	25.47	14.67	21.37	2.97	4.63	2.77	3.72	1.33	8.12	6.88	7.53	1.18	25.32	16.97	21.59	2.76
	D_BH/cm	53.34	25.36	41.04	10.37	23.34	6.58	15.15	5.23	16.84	11.13	14.29	4.14	32.44	19.84	24.02	7.97
LiDAR data (2016)	W_Cs-n/m	8.38	3.16	5.65	2.34	6.56	4.41	5.67	0.82	7.28	4.96	5.40	1.45	8.64	5.94	6.91	1.16
	W_Ce-w/m	7.62	3.02	5.32	2.15	8.12	4.49	6.79	1.15	4.96	2.95	3.69	0.82	7.22	4.76	6.18	0.85
	H/m	26.85	15.85	22.55	2.97	5.48	3.05	4.56	1.34	9.00	6.79	7.69	1.30	25.85	17.45	22.12	2.85
	D_BH/cm	54.73	26.72	42.42	9.76	24.32	7.12	16.69	5.59	19.12	12.75	16.09	3.95	33.72	21.00	25.13	7.51
样地测量（2017） Sample plot measurement (2017)	W_Cs-n/m	8.83	3.35	5.83	2.74	6.80	4.57	5.58	0.69	7.63	5.03	5.93	1.28	9.23	6.43	7.42	1.32
	W_Ce-w/m	8.22	3.21	5.29	2.43	8.31	4.72	6.91	1.08	5.22	3.02	4.01	0.76	7.82	5.28	6.76	0.93
	H/m	28.04	16.84	23.64	3.28	6.64	3.81	4.92	1.23	9.94	7.64	8.55	1.65	26.24	17.96	22.75	2.98
	D_BH/cm	56.07	27.97	43.67	9.79	25.97	8.17	18.06	4.73	21.47	14.17	17.79	3.54	35.17	22.08	25.68	6.54
LiDAR data (2018)	W_Cs-n/m	9.13	3.31	5.97	2.55	6.95	4.76	5.73	0.73	7.98	5.28	6.35	1.34	9.82	6.95	8.08	0.98
	W_Ce-w/m	8.47	3.24	5.41	2.31	8.34	4.88	7.02	1.25	5.49	3.12	4.06	0.84	8.38	5.87	7.17	0.91
	H/m	29.38	18.02	24.79	3.07	7.42	4.25	5.43	1.35	10.78	8.03	9.05	1.79	26.78	18.24	23.36	3.12
	D_BH/cm	57.34	29.18	44.98	9.45	27.48	8.98	19.54	5.26	23.11	14.06	18.37	4.24	36.98	23.24	26.86	6.90
样地测量（2019） Sample plot measurement (2019)	W_Cs-n/m	9.50	3.48	6.11	2.13	7.18	4.94	5.74	0.78	8.14	5.36	6.79	1.62	10.49	7.34	8.76	1.04
	W_Ce-w/m	8.93	3.29	5.85	1.96	8.63	5.03	7.13	1.27	5.63	3.15	4.37	0.72	8.83	6.23	7.63	0.97
	H/m	30.74	19.28	26.01	3.23	8.24	4.89	6.47	1.32	11.71	8.48	9.76	2.14	27.31	18.71	23.86	2.70
	D_BH/cm	58.92	30.69	46.18	10.29	29.05	9.55	20.68	5.18	25.41	14.97	18.92	3.40	37.51	23.65	27.47	7.14
LiDAR data (2020)	W_Cs-n/m	9.72	3.51	6.24	2.29	7.42	5.12	5.98	0.87	8.92	5.72	7.11	1.74	11.32	8.01	9.43	1.17
	W_Ce-w/m	9.66	3.35	5.98	2.07	8.87	5.22	7.39	1.36	5.96	3.31	4.63	0.81	9.42	6.82	8.12	0.97
	H/m	31.63	20.31	26.97	2.85	9.45	5.96	7.68	1.28	12.92	9.02	10.62	1.95	27.82	19.22	24.43	2.80
	D_BH/cm	60.06	31.65	47.06	9.20	30.65	10.35	21.09	5.71	27.65	16.15	20.68	3.79	38.76	24.07	28.53	6.95
样地测量（2021） Sample plot measurement (2021)	W_Cs-n/m	10.06	3.58	6.46	2.54	7.66	5.31	6.17	0.74	9.18	6.18	7.68	1.96	12.00	8.61	10.17	1.26
	W_Ce-w/m	9.48	3.32	6.45	2.36	9.07	5.32	7.65	1.21	6.14	3.47	4.74	0.98	10.16	7.38	8.62	0.88
	H/m	32.98	21.46	28.17	3.43	10.32	6.48	8.13	1.27	13.87	9.87	11.67	1.78	28.37	19.87	24.95	2.94
	D_BH/cm	61.17	32.87	48.13	9.63	32.01	11.12	22.13	5.03	29.91	17.41	22.46	3.97	39.41	24.41	29.36	7.31
样地测量（2022） Sample plot measurement (2022)	W_Cs-n/m	10.42	3.65	6.65	2.40	7.84	5.56	6.36	0.52	9.39	6.49	7.94	1.91	12.84	9.45	10.91	1.10
	W_Ce-w/m	9.89	3.37	6.42	2.17	9.34	5.25	7.78	1.04	6.35	3.65	4.84	0.86	10.81	8.09	9.36	0.97
	H/m	33.87	22.34	29.21	3.15	11.44	7.14	9.28	1.37	14.74	10.34	12.28	1.60	28.84	20.34	24.49	2.77
	D_BH/cm	62.33	33.95	49.08	9.86	33.05	11.95	23.34	5.35	31.92	18.45	24.77	4.19	40.32	24.7	30.44	7.02
注：W_Cs-n为南北方向冠宽；W_Ce-w为东西方向冠宽；H为树高；D_BH为胸径。Notes: W_Cs-n is crown width in north-south direction，W_Ce-w is crown width in east-west direction, H is tree height, and D_BH is DBH.

下载: 导出CSV

表 2 基于树龄的线性回归和基于LSTM的深度学习网络与本研究方法在相同测试数据中的生长参数预测精度比较

Table 2. Comparison of the accuracy of growth parameter prediction using linear regression based on tree age, deep learning network based on LSTM and our method on the same test data

方法 Method	胸径 DBH		树高 Tree height		冠宽 Crown width
方法 Method	R²	E_RMS/cm	R²	E_RMS/m	R²	E_RMS/m
基于树龄的线性回归^[24] Linear regression based on tree age	0.67	2.98	0.75	0.85	0.72	0.97
基于LSTM的深度学习网络^[14] Deep learning network based on LSTM	0.79	1.86	0.84	1.19	0.80	0.62
本研究方法 Our method	0.88	1.47	0.88	0.35	0.86	0.22
注：基于树龄的线性回归模型和基于LSTM的深度学习网络模型的预测精度数据，源自我们使用相同的训练集与测试集构建的线性回归与LSTM模型。Notes: the predictive accuracy of the age-based linear regression model and the LSTM-based deep learning network model is derived from the linear regression and LSTM models we built using the same training and test sets.

下载: 导出CSV

参考文献(31)

[1]	夏卫兵. 李树桥小流域湿地松生长规律及Verhaulst预测模型的建立与讨论[J]. 江西水利科技, 2022, 48(4): 285−290, 295. Xia W B. Establishment and discussion of Pinus elliottii growth law and verhaulst prediction model in Lishuqiao Small Watershed[J]. Jiangxi Hydraulic Science & Technology, 2022, 48(4): 285−290, 295.
[2]	洪李斌, 卿蕴贤, 田佳赫, 等. 基于混合效应模型的塞罕坝华北落叶松人工林单木去皮胸径生长预测[J]. 林业与生态科学, 2022, 37(2): 127−133. Hong L B, Qing Y X, Tian J H, et al. Prediction of single tree inside bark DBH growth ofLarix principis-rupprechtii plantation in Saihanba based on mixed effects model[J]. Forestry and Ecological Science, 2022, 37(2): 127−133.
[3]	Mohamedou C, Korhonen L, Eerikäinen K, et al. Using LiDAR-modified topographic wetness index, terrain attributes with leaf area index to improve a single-tree growth model in south-eastern Finland[J]. Forestry: an International Journal of Forest Research, 2019, 92(3): 253−263. doi: 10.1093/forestry/cpz010
[4]	郭常酉, 郭宏仙, 王宝华. 基于气候因子的杉木单木胸径生长模型构建[J]. 南京林业大学学报(自然科学版), 2023, 47(1): 47−56. Guo C Y, Guo H X, Wang B H. Study on incremnet model of individual-tree diameter of Cunninghamia lanceolata in consideration of climate factors[J]. Journal of Nanjing Forestry University (Natural Science Edition), 2023, 47(1): 47−56.
[5]	王君杰, 姜立春. 基于线性分位数组合的兴安落叶松冠幅预测[J]. 南京林业大学学报(自然科学版), 2021, 45(5): 161−170. Wang J J, Jiang L C. Predicting crown width for Larix gmelinii based on linear quantiles groups[J]. Journal of Nanjing Forestry University (Natural Science Edition), 2021, 45(5): 161−170.
[6]	林卓, 吴承祯, 洪伟, 等. 杉木人工林碳汇木材复合经济收益分析及最优轮伐期确定: 基于时间序列预测模型[J]. 林业科学, 2016, 52(10): 134−145. Lin Z, Wu C Z, Hong W, et al. Economic benefits analysis of carbon sequestration and timber and determination of optimal rotation period for a Cunninghamia lanceolata plantation: based on time series model[J]. Scientia Silvae Sinicae, 2016, 52(10): 134−145.
[7]	欧强新, 雷相东, 沈琛琛, 等. 基于随机森林算法的落叶松–云冷杉混交林单木胸径生长预测[J]. 北京林业大学学报, 2019, 41(9): 9−19. Ou Q X, Lei X D, Shen C C, et al. Individual tree DBH growth prediction of larch-spruce-fir mixed forests based on random forest algorithm[J]. Journal of Beijing Forestry University, 2019, 41(9): 9−19.
[8]	Liu H, Shen X, Cao L, et al. Deep learning in forest structural parameter estimation using airborne lidar data[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 1603−1618. doi: 10.1109/JSTARS.2020.3046053
[9]	da Silva A K V, Borges M V V, Batista T S, et al. Predicting eucalyptus diameter at breast height and total height with UAV-based spectral indices and machine learning[J]. Forests, 2021, 12(5): 582. doi: 10.3390/f12050582
[10]	Magalhaes J G S, Polinko A P, Amoroso M M, et al. The predicting tree growth app: an algorithmic approach to modelling individual tree growth[J]. Ecological Modelling, 2022, 467: 109932. doi: 10.1016/j.ecolmodel.2022.109932
[11]	Gong L, Yu M, Jiang S, et al. Deep learning based prediction on greenhouse crop yield combined TCN and RNN[J]. Sensors, 2021, 21(13): 4537. doi: 10.3390/s21134537
[12]	Sakurai S, Uchiyama H, Shimada A, et al. Plant growth prediction using convolutional LSTM[C]//Cláudio A P, Bouatouch K, Chessa M, et al. VISIGRAPP: international joint conference on computer vision, imaging and computer graphics. Prague: Springer International Publishing, 2019: 105−113.
[13]	Qiao M, He X, Cheng X, et al. KSTAGE: a knowledge-guided spatial-temporal attention graph learning network for crop yield prediction[J]. Information Sciences, 2023, 619: 19−37. doi: 10.1016/j.ins.2022.10.112
[14]	de Alwis S, Zhang Y, Na M, et al. Duo attention with deep learning on tomato yield prediction and factor interpretation[C]//Nayak A C, Sharma A.PRICAI 2019 trends in artificial intelligence: 16th pacific rim international conference on artificial intelligence. Cuvu: Springer International Publishing, 2019: 704−715.
[15]	Tian H, Wang P, Tansey K, et al. A deep learning framework under attention mechanism for wheat yield estimation using remotely sensed indices in the Guanzhong Plain, P.R. China[J]. International Journal of Applied Earth Observation and Geoinformation, 2021, 102: 102375. doi: 10.1016/j.jag.2021.102375
[16]	Han D, Wang P, Tansey K, et al. Integrating an attention-based deep learning framework and the SAFY-V model for winter wheat yield estimation using time series SAR and optical data[J]. Computers and Electronics in Agriculture, 2022, 201: 107334. doi: 10.1016/j.compag.2022.107334
[17]	Alhnaity B, Kollias S, Leontidis G, et al. An autoencoder wavelet based deep neural network with attention mechanism for multi-step prediction of plant growth[J]. Information Sciences, 2021, 560: 35−50. doi: 10.1016/j.ins.2021.01.037
[18]	Yang X, Yang H, Zhang F, et al. A random-weighted plane-Gaussian artificial neural network[J]. Neural Computing and Applications, 2019, 31: 8681−8692. doi: 10.1007/s00521-019-04457-6
[19]	Zhang W, Qi J, Wan P, et al. An easy-to-use airborne LiDAR data filtering method based on cloth simulation[J]. Remote Sensing, 2016, 8(6): 501. doi: 10.3390/rs8060501
[20]	Yun T, Jiang K, Li G, et al. Individual tree crown segmentation from airborne LiDAR data using a novel Gaussian filter and energy function minimization-based approach[J]. Remote Sensing of Environment, 2021, 256: 112307. doi: 10.1016/j.rse.2021.112307
[21]	Yun T, Jiang K, Hou H, et al. Rubber tree crown segmentation and property retrieval using ground-based mobile LiDAR after natural disturbances[J]. Remote Sensing, 2019, 11(8): 903. doi: 10.3390/rs11080903
[22]	张卫正, 刘晓燕, 王国飞, 等. 基于车载LiDAR的行道树胸径和株距测量[J]. 轻工学报, 2017, 32(3): 78−84. Zhang W Z, Liu X Y, Wang G F, et al. Measurement of DBH and row spacing of street trees based on mobile LiDAR[J]. Journal of Light Industry, 2017, 32(3): 78−84.
[23]	李丹, 庞勇, 岳彩荣, 等. 基于TLS数据的单木胸径和树高提取研究[J]. 北京林业大学学报, 2012, 34(4): 79−86. doi: 10.13332/j.1000-1522.2012.04.027 Li D, Pang Y, Yue C R, et al. Extraction of individual tree DBH and height based on terrestrial laser scanner data[J]. Journal of Beijing Forestry University, 2012, 34(4): 79−86. doi: 10.13332/j.1000-1522.2012.04.027
[24]	李智华, 杨小林, 杨雯婧,等. 巨柏胸径(d_1.3 < 1 m)和树龄的关系研究[J]. 高原农业, 2023, 7(3): 273−277. Li Z H, Yang X L, Yang W J, et al. Relationship between diamater at breast height (d_1.3 < 1 m) and age of Cupressus gigantea[J]. Journal of Plateau Agriculture, 2023, 7(3): 273−277.
[25]	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. Shen J B, Lei X D, Li Y T, et al. Prediction mean height for Larix olgensis plantation based on Bayesain-regularization BP neural network[J]. Journal of Nanjing Forestry University (Natural Science Edition), 2018, 42(2): 147−154.
[26]	单凯丽, 臧颢, 潘萍, 等. 赣中天然闽楠单木冠幅预测模型的研究[J]. 南京林业大学学报(自然科学版), 2022, 46(1): 88−96. Shan K L, Zang H, Pan P, et al. Individual tree crown width prediction models for natural Phoebe bournei in central Jiangxi Province[J]. Journal of Nanjing Forestry University (Natural Science Edition), 2022, 46(1): 88−96.
[27]	Yang Z, Liu Q, Luo P, et al. Prediction of individual tree diameter and height to crown base using nonlinear simultaneous regression and airborne LiDAR data[J]. Remote Sensing, 2020, 12(14): 2238. doi: 10.3390/rs12142238
[28]	Li Y, Zhu Z, Kong D, et al. EA-LSTM: evolutionary attention-based LSTM for time series prediction[J]. Knowledge-Based Systems, 2019, 181: 104785. doi: 10.1016/j.knosys.2019.05.028
[29]	Zheng W, Chen G. An accurate GRU-based power time-series prediction approach with selective state updating and stochastic optimization[J]. IEEE Transactions on Cybernetics, 2021, 52(12): 13902−13914.
[30]	Liu F T, Ting K M, Zhou Z H. Isolation forest[C]//2008 Eighth IEEE International Conference on data mining. New York: IEEE, 2008: 413−422.
[31]	Bazezew M N, Hussin Y A, Kloosterman E H. Integrating airborne LiDAR and terrestrial laser scanner forest parameters for accurate above-ground biomass/carbon estimation in Ayer Hitam tropical forest, Malaysia[J]. International Journal of Applied Earth Observation and Geoinformation, 2018, 73: 638−652. doi: 10.1016/j.jag.2018.07.026