联合LiDAR、高光谱数据及3D-CNN方法的树种分类

doi:10.11707/j.1001-7488.LYKX20220533

摘要/Abstract

摘要：

目的: 探究三维卷积神经网络（3D-CNN）在高光谱数据支持的树种分类中的有效网络构建方式，以提高树种分类精度。方法: 以美国加利福尼亚州内华达山脉南部为研究区，LiDAR数据获取的森林冠层高（CHM）进行单木分割并以此为补充建立样本，改进一种结构更简单、分类精度更高且无需对高光谱数据进行预处理的3D-CNN网络结构用于森林树种识别。结果: 相较于常规机器学习分类方法【支持向量机（SVM），随机森林（RF）】、传统二维卷积神经网络模型（2D-CNN）及最新多光谱分辨率三维卷积神经网络（MSR 3D-CNN）模型，本研究提出的3D-CNN模型对树种总体分类精度为99.79%，平均交并比（MIoU）为99.53%。与SVM和RF分类结果相比，本研究构建的3D-CNN模型总体分类精度提高5%左右，且具有对树种边界提取更加准确、椒盐现象更少发生的特点；与2D-CNN相比，总体分类精度提高10%左右，MIoU提高7%左右；与MSR 3D-CNN相比，总体精度相差不大，但在训练和测试过程中，本模型耗时远远小于MSR 3D-CNN模型。结论: 本研究改进的3D-CNN模型结构能够高效对原始高光谱影像进行树种分类并制图，可有效提高树种分类的精度。

关键词: 高光谱, LiDAR, 卷积神经网络, 树种分类, 3D-CNN

Abstract:

Objective: To explore the effective network construction method of three-dimensional convolutional neural network (3D-CNN) in tree species classification supported by hyperspectral data. Method: Taking the southern Sierra Nevada, California, USA as the study area, the canopy height model (CHM) obtained from LiDAR data was divided into single trees and used as a supplement to establish samples. A 3D-CNN network structure with simpler structure, higher classification accuracy and no need to preprocess hyperspectral data was improved for forest species identification. Result: Compared with the conventional supervised classification methods(support vector machine, random forest), the traditional two-dimensional convolutional neural network model and the latest MSR 3D-CNN model, the overall classification accuracy of the 3D-CNN model proposed in this study is 99.79%, and the mean intersection over union(MIoU) is 99.53%. Compared with SVM and RF method, the overall classification accuracy is improved by about 5%, and the new 3D-CNN model has the characteristics of more accurate extraction of tree species' boundaries and less pepper and salt phenomenon; Compared with 2D-CNN, the overall classification accuracy is improved by about 10%, and MIoU is improved by about 7%; Compared with MSR 3D-CNN, the overall accuracy is not much different, but in the process of training and testing, this model takes much less time than MSR 3D-CNN model. Conclusion: The 3D-CNN model proposed in this study can efficiently process the original hyperspectral images, classify and map tree species, and add forest vertical structure information to make classification labels, which can obtain higher accurate classification results.

Key words: hyperspectral, LiDAR, convolutional neural network, tree species classification, 3D-CNN

中图分类号:

S757

毛英伍,郭颖,张王菲,苏勇,关塬. 联合LiDAR、高光谱数据及3D-CNN方法的树种分类[J]. 林业科学, 2023, 59(3): 73-83.

Yingwu Mao,Ying Guo,Wangfei Zhang,Yong Su,Yuan Guan. Tree Species Classification by Combining LiDAR, Hyperspectral Data and 3D-CNN Method[J]. Scientia Silvae Sinicae, 2023, 59(3): 73-83.

图/表 12

图1

表1

图2

表2

图3

表3

图4

表4

图5

图6

图7

图8

参考文献 0

	刘怡君, 庞　勇, 廖声熙, 等. 2016. 机载LiDAR和高光谱融合实现普洱山区树种分类. 林业科学研究, 29(3): 407-412.
	Liu Y J, Pang Y, Liao S X, et al. 2016. Merged airborne LiDAR and hyperspectral data for tree species classification in Pu'er mountain area. Forest Research, 29 (3): 407-412. ［in Chinese］
	卢元兵, 李华朋, 张树清. 2021. 基于混合3D-2D CNN的多时相遥感农作物分类. 农业工程学报, 37(13): 142-151.
	Lu Y B, Li H P, Zhang S Q. 2021. Multi-temporal remote sensing based crop classification using a hybrid 3D-2D CNN model. Transactions of the Chinese Society of Agricultural Engineering, 37(13): 142-151. ［in Chinese］
	王　彬. 2018. 基于机载LiDAR和高光谱数据的单木提取和树种识别. 南京: 南京信息工程大学.
	Wang B. 2018. Individual tree extraction and species identification based on airborne LiDAR and hyperspectral data. Nanjing: Nanjing University of Information Engineering. ［in Chinese］
	赵　霖. 2019. 基于机载高光谱数据空谱联合特征的3D-CNN树种分类算法. 北京: 北京林业大学, 79.
	Zhao L 2019. Three-dimensional convolutional neuarl network algorithm for forest tree species classification using airborne hyperspectral spatial-spectral joint features. Beijing: Beijing Forestry University, 79. ［in Chinese］
	赵　霖, 张晓丽, 吴艳双, 等. 2020. 面向机载高光谱数据的3D-CNN亚热带森林树种分类. 林业科学, 56(11): 97−107.
	Zhao L, Zhang X L, Wu Y S, et al 2020. Subtropical forest tree species classification based on 3D-CNN for airborne hyperspectral data. Scientia Silvae Sinicae, 56 (11): 97-107. ［in Chinese］
	Laurel B, Leonhard B, Ellen H, et al Tree species classification using hyperspectral imagery: a comparison of two classifiers. Remote Sensing, 2016, 8 (6): 445.
	Dalponte M, Frizzera L, Gianelle D Individual tree crown delineation and tree species classification with hyperspectral and LiDAR data. PeerJ, 2019, 6, e6227. doi: 10.7717/peerj.6227
	Fassnacht F E, Latifi H, Stereńczak K, et al Review of studies on tree species classification from remotely sensed data. Remote Sensing of Environment, 2016, 186, 64- 87. doi: 10.1016/j.rse.2016.08.013
	Fricker G A, Ventura J D, Wolf J A, et al A convolutional neural network classifier identifies tree species in mixed-conifer forest from hyperspectral imagery. Remote Sensing, 2019, 11 (19): 2326. doi: 10.3390/rs11192326
	Hinton G E, Srivastava N, Krizhevsky A, et al. 2013. Improving neural networks by preventing co-adaptation of feature detectors. Neural and Evolutionary Computing, arXiv:1207.0580.
	Hinton G, Osindero S, Welling M, et al Unsupervised discovery of non-linear structure using contrastive backpropagation. Cognitive Science, 2008, 30 (4): 725- 731.
	Kingma D, Ba J. 2014. Adam: a method for stochastic optimization. Computer Science, arXiv:1412.6980.
	Krizhevsky A, Sutskever I, Hinton G ImageNet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 2012, 25 (2): 1097- 1105.
	Li Y, Zhang H, Shen Q Spectral–Spatial classification of hyperspectral imagery with 3D convolutional neural network. Remote Sensing, 2017, 9 (1): 67.
	Marrs J, Ni-meister W Machine learning techniques for tree species classification using co-registered LiDAR and hyperspectral data. Remote Sensing, 2019, 11 (7): 819. doi: 10.3390/rs11070819
	Maschler J, Atzberger C, Immitzer M Individual tree crown segmentation and classification of 13 tree species using airborne hyperspectral data. Remote Sensing, 2018, 10 (8): 1218. doi: 10.3390/rs10081218
	Mäyrä J, Keski-saari S, Kivinen S, et al Tree species classification from airborne hyperspectral and LiDAR data using 3D convolutional neural networks. Remote Sensing of Environment, 2021, 256, 112322. doi: 10.1016/j.rse.2021.112322
	Meyer M D, North M P, Gray A N, et al Influence of soil thickness on stand characteristics in a Sierra Nevada mixed-conifer forest. Plant & Soil, 2007, 294 (1/2): 113- 123.
	Modzelewska A, Fassnacht F E, Stereńczak K Tree species identification within an extensive forest area with diverse management regimes using airborne hyperspectral data. International Journal of Applied Earth Observation and Geoinformation, 2020, 84, 101960. doi: 10.1016/j.jag.2019.101960
	North M, Oakley B, Chen J, et al. 2002. Vegetation and ecological characteristics of mixed-conifer and red fir forests at the teakettle experimental forest. General Technical Report PSW-GTR-186.
	Sankey T, Donager J, Mcvay J, et al UAV LiDAR and hyperspectral fusion for forest monitoring in the southwestern USA. Remote Sensing of Environment, 2017, 195, 30- 43. doi: 10.1016/j.rse.2017.04.007
	Tran D, Bourdev L, Fergus R, et al. 2015. Learning spatiotemporal features with 3D convolutional networks. IEEE, arXiv:1412.0767.
	Xu H, Yao W, Cheng L, et al Multiple spectral resolution 3D convolutional neural network for hyperspectral image classification. Remote Sensing, 2021, 13 (7): 1248. doi: 10.3390/rs13071248
	Zhang B, Zhao L, Zhang X Three-dimensional convolutional neural network model for tree species classification using airborne hyperspectral images. Remote Sensing of Environment, 2020, 247, 111938. doi: 10.1016/j.rse.2020.111938

编码Code	分类名称Name	数量Number
1	红冷杉Abies magnifica	8
2	北美翠柏Calocedrus decurrens	29
3	加州黄松Pinus jeffreyi	38
4	糖松Pinus lambertiana	16
5	加州黑栎Quercus kelloggii	8
6	扭叶松 Pinus contorta	5
7	死树Any species	28
8	科罗拉多冷杉 Abies concolor	16
9	裸地Land	69

网络层Layer	卷积核大小Filter Size	步长Stride	输出大小Output size	激活函数Activation	丢弃率Dropout
Input	—		13×13×426	—	—
conv3D	3×3×7,8	1×1×1	11×11×420	ReLU	—
conv3D	3×3×5,16	1×1×1	9×9×416	ReLU	—
conv3D	3×3×3,32	1×1×1	7×7×414	ReLU	—
conv3D	3×3×3,64	1×1×1	5×5×414	ReLU	—
FC	256	—	1×256	ReLU	0.4
FC	128	—	1×128	ReLU	0.4
Softmax	9	—	1×9	Softmax	—

项目 Item	RF	SVM	2D-CNN	MSR 3D-CNN	3D-CNN
OA(%)	94.41	96.16	88.88	99.76	99.79
AA(%)	62.62	73.78	92.75	94.64	95.56
MIoU(%)	—	—	92.43	97.70	99.53
训练时间Training time/s	360	780	796.71	254539	32808
测试时间Test time/s	43	62	82	322	244

树种Species	精度Precision	召回率Recall	F1分数F1-score	数量Support
科罗拉多冷杉 Abies concolor	1.00	1.00	1.00	28
红冷杉Abies magnifica	1.00	0.99	0.99	175
北美翠柏Calocedrus decurrens	1.00	1.00	1.00	296
加州黄松Pinus jeffreyi	0.99	1.00	0.99	142
糖松Pinus lambertiana	1.00	0.57	0.73	7
加州黑栎Quercus kelloggii	1.00	1.00	1.00	12
扭叶松 Pinus contorta	0.98	1.00	0.99	138
死树 Any species	1.00	1.00	1.00	90
裸地Land	1.00	1.00	1.00	802
精度Accuracy			1.00	1690
宏平均值Macro avg.	1.00	0.96	0.97	1690
加权平均值weighted avg.	1.00	1.00	1.00	1690

[1]	冯慧敏,金崑. 基于卷积神经网络的雄性海南长臂猿声纹识别[J]. 林业科学, 2023, 59(1): 119-127.
[2]	朱泊东,罗洪斌,金京,岳彩荣. 高郁闭度人工林无人机激光雷达单木分割方法优化[J]. 林业科学, 2022, 58(9): 48-59.
[3]	齐建东,谭新新. 长白山红松阔叶林的净碳交换变化及基于时间卷积神经网络的模拟[J]. 林业科学, 2022, 58(2): 1-12.
[4]	冯林艳,谭炳香,刘清旺,周超凡,于航,张会儒,符利勇. 基于GF-2影像的崇礼冬奥核心区土地覆盖和树种分类[J]. 林业科学, 2022, 58(10): 10-23.
[5]	谢栋博,刘清旺,雷雅凯,于航,杨旭平,符利勇. 基于机载LiDAR数据的崇礼冬奥核心区单木分割算法适宜性分析[J]. 林业科学, 2022, 58(10): 121-130.
[6]	何拓,刘守佳,陆杨,张永刚,焦立超,殷亚方. iWood: 基于卷积神经网络的濒危珍贵树种木材自动识别系统[J]. 林业科学, 2021, 57(9): 152-159.
[7]	赵子宇,杨霄霞,郭慧,葛浙东,周玉成. 基于卷积神经网络模型的木材宏、微观辨识方法[J]. 林业科学, 2021, 57(6): 134-143.
[8]	由珈齐,李明泽,范文义,全迎,王斌,莫祝坤,祝子枭. 基于高光谱和激光雷达数据的林分类型识别[J]. 林业科学, 2021, 57(5): 119-129.
[9]	霍朗宁,张晓丽. 点云数据双向选择单木提取与地面数据匹配方法[J]. 林业科学, 2021, 57(3): 181-188.
[10]	宿恒硕,吕军,丁志平,唐彦杰,陈旭东,周强,张哲宇,姚青. 基于改进残差神经网络的木材识别算法[J]. 林业科学, 2021, 57(12): 147-154.
[11]	刘璇昕,孙钰,崔剑,蒋琦,陈志泊,骆有庆. 钻蛀性害虫取食声音的人工智能早期识别[J]. 林业科学, 2021, 57(10): 93-101.
[12]	霍朗宁,张晓丽. 基于机载LiDAR点云多层聚类的单木信息提取及其精度评价[J]. 林业科学, 2021, 57(1): 85-94.
[13]	谢锦莹,丁丽霞,王志辉,刘丽娟. 基于FCN与面向对象的滨海湿地植被分类[J]. 林业科学, 2020, 56(8): 98-106.
[14]	郭颖,李增元,陈尔学,张旭,赵磊,陈艳,王雅慧. 一种改进的高空间分辨率遥感影像森林类型深度学习精细分类方法:双支FCN-8s[J]. 林业科学, 2020, 56(3): 48-60.
[15]	赵霖,张晓丽,吴艳双,张斌. 面向机载高光谱数据的3D-CNN亚热带森林树种分类[J]. 林业科学, 2020, 56(11): 97-107.