基于多分类器集成的落叶松人工林提取

doi:10.11707/j.1001-7488.20211111

摘要/Abstract

摘要：

目的: 探讨多时相、多光谱和高空间分辨率影像在落叶松人工林识别中的应用潜力，通过多种特征组合方案，寻找一种多分类器集成的落叶松人工林快速识别方法，为落叶松人工林后续监测与管理提供参考。方法: 以黑龙江省孟家岗林场为研究区，基于Landsat8 OLI影像分析不同物候期树种间的光谱差异，确定落叶松人工林识别的关键波谱和物候期，同时提取多种特征信息，通过变量重要性（VIM）筛选并构建不同物候相的多特征数据集。综合随机森林（RF）、支持向量机（SVM）、最大似然（MLC）和BP神经网络4种分类算法优势，设计一种多分类器集成的分类策略进行落叶松人工林提取。结果: 在多分类器集成的分类策略下，分类总精度达93.8%，面积提取精度达96.3%；与RF、MLC、SVM和BP等分类算法相比，多分类器集成分类策略的平均分类精度提高10%。结论: 相比单一时相影像，多时相影像数据包含落叶松人工林更多物候期，可反映出落叶松人工林独特的季相特征，有利于落叶松人工林识别。多分类器集成策略综合各分类器优点，可有效提高分类精度，实现落叶松人工林高精度提取。

关键词: 落叶松人工林, 森林类型分类, GF-1, 多分类器集成

Abstract:

Objective: This study was implemented to explore the potential of multi-temporal, multi-spectral, and high-spatial-resolution images in larch plantations identification, and to find a method for rapid identification of larch plantations integrated with multiple classifiers by multi-feature combination scheme, so as to provide a reference for remote sensing monitoring and management of larch plantations and forest resource investigation. Method: In this paper, first of all, the key period and parameters of larch identification were determined from multi-temporal Landsat8 OLI images of Mengjiagang forest farm in Heilongjiang Province by analyzing the seasonal and spectral features of larch. Secondly, the feature information extracted from images was screened through the variable importance metric (VIM) to establish multi-feature data sets of different phenological stages. Finally, a classification method with the multi-classifier ensemble was established by using the advantages of random forest (RF), support vector machine (SVM), maximum likelihood classification (MLC), and back-propagation (BP) neural network, to extract larch plantations. Result: Using the classification of multi-classifier integration, the overall accuracy was 93.8%, and the accuracy of extracted larch plantation area was 96.3%. Compared with RF, MLC, SVM and BP, the overall accuracy was improved by 10% on average. Conclusion: This paper found that compared with single-phase images, multi-temporal data contains more phenological phases of larch, reflecting the unique seasonal characteristics of larch, which is more conducive to the extraction of larch. At the same time, the multi-classifier integration strategy combines the advantages of each classifier, which would effectively improve the overall accuracy and achieve high-precision extraction of larch plantation.

Key words: larch plantation, classification of forest types, GF-1, multiple classifier integration

中图分类号:

S757.3

马婷,李崇贵,汤伏全,吕杰. 基于多分类器集成的落叶松人工林提取[J]. 林业科学, 2021, 57(11): 105-118.

Ting Ma,Chonggui Li,Fuquan Tang,Jie Lü. Extraction of Larch Plantation Based on Multi-Classifier Ensemble[J]. Scientia Silvae Sinicae, 2021, 57(11): 105-118.

图/表 13

表1

图1

表2

表3

图2

图3

图4

表4

表5

表6

图5

图6

图7

参考文献 0

	陈春秀, 陈蜀江, 徐世薇, 等. 多特征辅助下的GF-6 WFV影像准噶尔山楂识别研究. 干旱区研究, 2021, 38 (2): 553- 561.
	Chen C X , Chen S J , Xu S W , et al. Crataegus songarica recognition using Gaofen-6 wide-field-view data assisted by multiple features. Arid Zone Research, 2021, 38 (2): 553- 561.
	郭瑞霞, 李崇贵, 刘思涵, 等. 利用多时相特征的落叶松人工林分类. 浙江农林大学学报, 2020, 37 (2): 235- 242.
	Guo R X , Li C G , Liu S H , et al. Classification of Larix gmelinii plantation based on multi-temporal characteristics. Journal of Zhejiang A & F University, 2020, 37 (2): 235- 242.
	郭颖, 李增元, 陈尔学, 等. 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. 林业科学, 2020, 56 (3): 48- 60.
	Guo Y , Li Z Y , Chen E X , et al. A deep learning method for forest fine classification based on high resolution remote sensing images: two-branch FCN-8s. Scientia Silvae Sinicae, 2020, 56 (3): 48- 60.
	郭振华, 陈立红. 阿尔山不同时期落叶松根际土壤固氮菌的多样性研究. 西北植物学报, 2019, 39 (4): 729- 739.
	Guo Z H , Chen L H . Community structure and diversity of culturable nitrogen fixing bacteria from Larix gmelinii rhizosphere soil at different stages in Arxan. Acta Botanica Boreali-Occidentalia Sinica, 2019, 39 (4): 729- 739.
	郝睿, 徐俊峰, 王庆宝, 等. 基于BP神经网络的多特征融合变化检测方法. 海洋测绘, 2016, 36 (1): 79- 82. doi: 10.3969/j.issn.1671-3044.2016.01.020
	Hao R , Xun J F , Wang Q B , et al. Change detection method using multi-feature fusion based on BP neural network. Hydrographic Surveying and Charting, 2016, 36 (1): 79- 82. doi: 10.3969/j.issn.1671-3044.2016.01.020
	李春干, 邵国凡. 面向对象森林分类的多分类器结合方法研究. 南京林业大学学报(自然科学版), 2010, 34 (1): 73- 79. doi: 10.3969/j.issn.1000-2006.2010.01.016
	Li C G , Shao G F . Combination multiclassifier for object-oriented classification of forest cover. Journal of Nanjing Forestry University(Natural Science Edition), 2010, 34 (1): 73- 79. doi: 10.3969/j.issn.1000-2006.2010.01.016
	李敏, 李华, 程茂华. 一种自适应子融合集成多分类器方法. 计算机测量与控制, 2019, 27 (4): 120- 123.
	Li M , Li H , Chen M H . An adaptive sub-fusion integration classification method. Computer Measurement & Control, 2019, 27 (4): 120- 123.
	李小梅, 张秋良, 李增元, 等. 基于对象的CHRIS遥感图像森林类型分类方法研究. 内蒙古农业大学学报(自然科学版), 2010, 31 (2): 31- 36.
	Li X M , Zhang Q L , Li Z Y , et al. The study on the forest types classification method of CHRIS remote sensing image based on object. Journal of Inner Mongolia Agricultural University(Natural Science Edition), 2010, 31 (2): 31- 36.
	栗旭升, 李虎, 陈冬花, 等. 联合GF-5与GF-6卫星数据的多分类器组合亚热带树种识别. 林业科学, 2020, 56 (10): 93- 104. doi: 10.11707/j.1001-7488.20201010
	Li X S , Li H , Chen D H , et al. Multiple classifiers combination method for tree species identification based on GF-5 and GF-6. Scientia Silvae Sinicae, 2020, 56 (10): 93- 104. doi: 10.11707/j.1001-7488.20201010
	梁守真, 陈劲松, 吴炳方, 等. 应用面向对象的决策树模型提取橡胶林信息. 遥感学报, 2015, 19 (3): 485- 494.
	Liang S Z , Chen J S , Wu B F , et al. Extracting rubber plantation decision tree model based on object-oriented method. National Remote Sensing Bulletin, 2015, 19 (3): 485- 494.
	刘林. 我国森林可持续利用的生态经济弹性分析. 林业科学, 2013, 49 (8): 132- 139.
	Liu L . Analysis of forest sustainable utilization in China using an ecological economy elasticity model. Scientia Silvae Sinicae, 2013, 49 (8): 132- 139.
	刘鲁霞, 庞勇, 任海保, 等. 基于高分2号遥感数据估测中亚热带天然林木本植物物种多样性. 林业科学, 2019, 55 (2): 61- 74.
	Liu L X , Pang Y , Reng H B , et al. Predict tree species diversity from GF-2 satellite data in a subtropical forest of China. Scientia Silvae Sinicae, 2019, 55 (2): 61- 74.
	刘向东, 骆斌, 陈兆乾. 支持向量机最优模型选择的研究. 计算机研究与发展, 2005, 48 (4): 576- 581.
	Liu X D , Luo B , Chen Z Q . Optimal model selection for support vector machines. Journal of Computer Research and Development, 2005, 48 (4): 576- 581.
	刘怡君, 庞勇, 廖声熙, 等. 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, 2016, 29 (3): 407- 412. doi: 10.3969/j.issn.1001-1498.2016.03.015
	Liu Y J , Pang Y , Liao S X , et al. Merged airborne LiDAR and hyperspectral data for tree species classification in Pu'er Mountain area. Forest Research, 2016, 29 (3): 407- 412. doi: 10.3969/j.issn.1001-1498.2016.03.015
	马婷, 李崇贵, 郭瑞霞, 等. 运用植被指数时序特征对落叶松人工林分类. 东北林业大学学报, 2020, 48 (3): 56- 59. doi: 10.3969/j.issn.1000-5382.2020.03.010
	Ma T , Li C G , Guo R X , et al. Larch extraction by vegetation index time series characteristics. Journal of Northeast Forestry University, 2020, 48 (3): 56- 59. doi: 10.3969/j.issn.1000-5382.2020.03.010
	戚王月, 胡宏祥, 夏萍, 等. 基于改进BP神经网络的多分辨率遥感图像分类及对比分析. 安徽农业大学学报, 2019, 46 (4): 737- 744.
	Qi W Y , Hu H X , Xia P , et al. Multi-resolution remote sensing images classification and comparative analysis based on improved BP neural network. Journal of Anhui Agricultural University, 2019, 46 (4): 737- 744.
	王怀警, 谭炳香, 王晓慧, 等. 多分类器组合森林类型精细分类. 遥感信息, 2019, 34 (2): 104- 112. doi: 10.3969/j.issn.1000-3177.2019.02.017
	Wang H J , Tan B X , Wang X H , et al. Multiple classifiers combination method for precise classification of forest type. Remote Sensing Information, 2019, 34 (2): 104- 112. doi: 10.3969/j.issn.1000-3177.2019.02.017
	王宏涛, 孙剑伟. 基于BP神经网络和SVM的分类方法研究. 软件, 2015, 36 (11): 96- 99. doi: 10.3969/j.issn.1003-6970.2015.11.024
	Wang H T , Sun J W . Research on the classification method based on BP neural network and SVM. Computer Engineering & Software, 2015, 36 (11): 96- 99. doi: 10.3969/j.issn.1003-6970.2015.11.024
	王磊, 贾炜玮, 李凤日. 孟家岗林场森林景观结构动态分析. 植物研究, 2008, 28 (4): 497- 502.
	Wang L , Jia W W , Li F R . The analysis of the dynamics of the landscape structure of Mengjiagang forest farm. Bulletin of Botanical Research, 2008, 28 (4): 497- 502.
	王健峰, 张磊, 陈国兴, 等. 基于改进的网格搜索法的SVM参数优化. 应用科技, 2012, 39 (3): 28- 31.
	Wang J F , Zhang L , Chen G X , et al. A parameter optimization method for an SVM based on improved grid search. Applied Science and Technology, 2012, 39 (3): 28- 31.
	王全才. 2011. 随机森林特征选择. 大连: 大连理工大学.
	Wang Q C. 2011. Random forest feature selection. Dalian: Dalian University of Technology. [in Chinese]
	王霓虹, 王志芳. 孟家岗落叶松人工林标准树高模型的研究. 森林工程, 2014, 30 (2): 10- 12. doi: 10.3969/j.issn.1001-005X.2014.02.003
	Wang N H , Wang Z F . Study on standard tree height model of Mengjiagang's larch plantation. Forest Engineering, 2014, 30 (2): 10- 12. doi: 10.3969/j.issn.1001-005X.2014.02.003
	王建国, 陈肖洁, 张文兴. 支持向量回归机核参数优化研究. 内蒙古科技大学学报, 2016, 35 (3): 255- 259.
	Wang J G , Chen X J , Zhang W X . Research on kernel parameter optimization of support vector regression. Journal of Inner Mongolia University of Science and Technology, 2016, 35 (3): 255- 259.
	项铭涛, 卫炜, 吴文斌. 植被物候参数遥感提取研究进展评述. 中国农业信息, 2018, 30 (1): 55- 66.
	Xiang M T , Wei W , Wu W B . Review of vegetation phenology estimation by using remote sensing. China Agricultural Informatics, 2018, 30 (1): 55- 66.
	徐晓明. 2014. SVM参数寻优及其在分类中的应用. 大连: 大连海事大学.
	Xu X M. 2014. SVM parameter optimization and its application in the classification. Dalian: Dalian Maritime University. [in Chinese]
	张茂震, 王广兴, 周国模, 等. 基于森林资源清查、卫星影像数据与随机协同模拟尺度转换方法的森林碳制图. 生态学报, 2009, 29 (6): 2919- 2928. doi: 10.3321/j.issn:1000-0933.2009.06.019
	Zhang M Z , Wang G X , Zhou G M , et al. Mapping of forest carbon by combining forest inventory data and satellite images with co-simulation based up-scaling method. Acta Ecologica Sinica, 2009, 29 (6): 2919- 2928. doi: 10.3321/j.issn:1000-0933.2009.06.019
	张晓羽, 李凤日, 甄贞, 等. 基于随机森林模型的陆地卫星-8遥感影像森林植被分类. 东北林业大学学报, 2016, 44 (6): 53- 57. doi: 10.3969/j.issn.1000-5382.2016.06.014
	Zhang X Y , Li F R , Zhen Z , et al. Forest vegetation classification of landsat8 remote sensing image based on random forests model. Journal of Northeast Forestry University, 2016, 44 (6): 53- 57. doi: 10.3969/j.issn.1000-5382.2016.06.014
	周宽久, 张世荣. 支持向量机分类算法研究. 计算机工程与应用, 2009, 45 (1): 159- 162, 182. doi: 10.3778/j.issn.1002-8331.2009.01.050
	Zhou K J , Zhang S R . Support vector machines based classification algorithm. Computer Engineering and Applications, 2009, 45 (1): 159- 162, 182. doi: 10.3778/j.issn.1002-8331.2009.01.050
	周星, 丁立新, 万润泽, 等. 分类器集成算法研究. 武汉大学学报(理学版), 2015, 61 (6): 503- 508.
	Zhou X , Ding L X , Wan R Z , et al. Research on classifier ensemble algorithms. Journal of Wuhan University(Natural Science Edition), 2015, 61 (6): 503- 508.
	Baird S R. 2001. Estimating mixtures of exponential distributions using maximum likelihood and the EM algorithm to improve simulation of telecommunications networks. British: The University of British Colunbia.
	Belgiu M , Drǎgut L . Random forest in remote sensing: a review of applications and future directions. ISPRS Journal of Photogrammetry and Remote Sensing, 2016, 114 (1): 24- 31.
	Bolyn C , Michez A , Gaucher P , et al. Forest mapping and species composition using supervised per pixel classification of Sentinel-2 imagery. Biotechnology, Agronomy and Society and Environment, 2018, 22 (2): 172- 190.
	Breiman L . Random forests. Machine Learning, 2001, 45 (1): 5- 32. doi: 10.1023/A:1010933404324
	Busetto L , Colombo R , Migliavacca M , et al. Remote sensing of larch phenological cycle and analysis of relationships with climate in the Alpine region. Global Change Biology, 2010, 16 (9): 1365- 2486.
	Cutler A , Cutler D R , Stevens J R . Random forests. Machine Learning, 2004, 45 (1): 157- 176.
	Evans J S , Cushman S A . Gradient modeling of conifer species using random forests. Landscape Ecology, 2009, 24 (5): 673- 683. doi: 10.1007/s10980-009-9341-0
	Immitzer M , Neuwirth M , Böck S , et al. Optimal input features for tree species classification in central Europe based on multi-temporal Sentinel-2 data. Remote Sensing, 2019, 11 (22): 2599- 2622. doi: 10.3390/rs11222599
	Kong F J , Li X B , Wang H , et al. Land cover classification based on fused data from GF-1 and MODIS NDVI time series. Remote Sensing, 2016, 8 (9): 741- 752. doi: 10.3390/rs8090741
	Li H , Chen Z X , Jiang Z W , et al. Comparative analysis of GF-1, HJ-1, and Landsat-8 data for estimating the leaf area index of winter wheat. Journal of Integrative Agriculture, 2017, 16 (2): 266- 285. doi: 10.1016/S2095-3119(15)61293-X
	Liu Y , Wang J B , Dong J W , et al. Variations of vegetation phenology extracted from remote sensing data over the Tibetan Plateau Hinterland during 2000-2014. Journal of Meteorological Research, 2020, 34 (4): 786- 797.
	Meng J , Li S , Wang W , et al. Estimation of forest structural diversity using the spectral and textural information derived from SPOT-5 satellite images. Remote Sensing, 2016, 8 (2): 1- 24.
	Nelson M. 2017. Evaluating multitemporal Sentinel-2 data for forest mapping using random forest. Stockholm: Stockholm University.
	Ouali M , King R D . Cascaded multiple classifiers for secondary structure prediction. Protein Science, 2010, 9 (6): 1162- 1176.
	Roy D P , Wulder M A , Loveland T R , et al. Landsat-8:science and product vision for terrestrial global change research. Remote Sensing of Environment, 2014, 15 (1): 68- 73.
	Statnikov A , Wang L , Aliferis C F . A comprehensive comparison of random forests and support vector machines for microarray-based cancer classification. BMC Bioinformatics, 2008, 9 (1): 319- 329. doi: 10.1186/1471-2105-9-319
	Strobl C , Boulesteix A L , Kneib T , et al. Conditional variable importance for random forests. BMC Bioinformatics, 2008, 9 (1): 307.
	Yu X W , Hyyppa J , Litkey P , et al. Single-sensor solution to tree species classification using multispectral airborne laser scanning. Remote Sensing, 2017, 9 (2): 108- 123.
	Zhu L , Li J , La Y X , et al. Improving the accuracy of remote sensing land cover classification by GEO-ECO zoning coupled with geostatistical simulation. Applied Sciences, 2021, 11 (2): 553- 568.

影像Image	产品号Product ID	成像时间Imaging date	条带号Path	行编号Row	幅宽Width/km²	分辨率Resolution/m
Landsat8 OLI	LC81150282016005LGN00	2016-01-04	115	28	185×170	15/30
	LC81150282018010LGN00	2018-01-10
	LC81150282018026LGN00	2018-01-26
	LC81150282018042LGN00	2018-02-11
	LC81150282018074LGN00	2018-03-15
	LC81150282018090LGN00	2018-03-31
	LC81150282018106LGN00	2018-04-16
	LC81150282016133LGN01	2016-05-12
	LC81150282017167LGN00	2017-06-16
	LC81150282017199LGN00	2017-07-18
	LC81150282017231LGN00	2017-08-19
	LC81150282017279LGN00	2017-10-06
	LC81150282017295LGN00	2017-10-22
	LC81150282015306LGN02	2015-11-02
	LC81150282017343LGN01	2017-12-09
	LC81150282017359LGN00	2017-12-25
GF-1 PMS1	2338810	2017-05-02	586	76	60×60	2/8
GF-1 PMS1	2338811	2017-05-02	586	77	60×60	2/8

数据集Data set	分类方案Classification scheme	属性类别Type of feature
单一特征影像 Single feature image	生长期Growth period	4月Landsat8 OLI影像光谱波段Spectral bands of Landsat8 OLI images in April
	生长旺盛期Vigorous growth period	6月Landsat8 OLI影像光谱波段Spectral bands of Landsat8 OLI images in June
	落叶期Deciduous period	10月Landsat8 OLI影像光谱波段Spectral bands of Landsat8 OLI images in October
	多时相Multi-temporal	4月+10月Landsat8 OLI影像光谱波段Spectral bands of Landsat8 OLI images in April & October
NIR-SWIR多特征数据集 NIR-SWIR multi-feature data	生长期Growth period	4月Landsat8影像中的NIR、SWIR1、SWIR2波段+VIM筛选的特征集+形状因子+面积因子NIR, SWIR1, SWIR2 bands in Landsat8 images in April + VIM screening feature set + shape factor + area factor
	生长旺盛期Vigorous growth period	6月Landsat8影像中的NIR、SWIR1、SWIR2波段+VIM筛选的特征集+形状因子+面积因子NIR, SWIR1, SWIR2 bands in Landsat8 images in June + VIM screening feature set + shape factor + area factor
	落叶期Deciduous period	10月Landsat8影像中的NIR、SWIR1、SWIR2波段+VIM筛选的特征集+形状因子+面积因子NIR, SWIR1, SWIR2 bands in Landsat8 images in October + VIM screening feature set + shape factor + area factor
	多时相Multi-temporal	4月Landsat8影像中的NIR、SWIR1、SWIR2波段+10月Landsat8影像中的NIR、SWIR1、SWIR2波段+VIM筛选的特征集+形状因子+面积因子NIR, SWIR1, SWIR2 bands in Landsat8 images in April + NIR, SWIR1, SWIR2 bands in Landsat8 images in October + VIM screening feature set + shape factor + area factor

子分类器 Sub-classifier	生长期 Growth period(%)					生长旺盛期 Vigorous growth period(%)					落叶期 Deciduous period(%)					多时相 Multi-temporal(%)
子分类器 Sub-classifier	PA	UA	WC	OA	Kappa	PA	UA	WC	OA	Kappa	PA	UA	WC	OA	Kappa	PA	UA	WC	OA	Kappa
MLC	83.1	97.4	26.5	80.5	0.70	66.9	89.7	23.8	70.1	0.62	84.5	96.1	25.7	81.4	0.71	85.5	97.2	26.8	83.5	0.72
RF	77.1	93.6	24.8	79.1	0.70	75.5	88.3	24.9	73.6	0.64	79.3	93.6	24.6	80.5	0.70	80.4	95.6	23.5	82.1	0.71
SVM	70.5	88.3	23.1	73.2	0.64	76.3	89.9	25.3	74.5	0.65	73.2	90.3	23.3	76.1	0.67	79.3	94.5	22.2	79.3	0.70
BP	80.3	95.7	25.6	80.3	0.70	79.5	91.6	26.0	78.6	0.69	87.1	98.5	26.4	82.9	0.71	86.1	96.9	27.5	86.5	0.73

子分类器 Sub-classifier	生长期 Growth period(%)					生长旺盛期 Vigorous growth period(%)					落叶期 Deciduous period(%)					多时相 Multi-temporal(%)
子分类器 Sub-classifier	PA	UA	WC	OA	Kappa	PA	UA	WC	OA	Kappa	PA	UA	WC	OA	Kappa	PA	UA	WC	OA	Kappa
MLC	95.4	99.9	27.3	82.1	0.71	70.8	82.3	22.1	73.1	0.64	96.1	99.8	28.5	84.0	0.72	92.4	100.0	24.2	84.1	0.72
RF	81.1	91.1	23.2	80.3	0.70	79.2	91.9	24.7	73.7	0.64	81.4	89.4	22.5	81.3	0.71	94.7	99.2	25.3	83.5	0.72
SVM	80.3	90.3	23.0	78.1	0.69	81.2	94.2	25.3	76.2	0.67	86.1	94.6	23.8	78.7	0.69	96.2	100.0	25.8	81.3	0.71
BP	92.6	99.8	26.5	81.2	0.71	89.5	99.8	27.9	78.8	0.69	94.8	99.9	26.2	83.9	0.72	91.7	99.2	24.7	87.1	0.74

指标 Indicators	单一特征影像Single feature data				NIR-SWIR多特征数据集NIR-SWIR multi-feature data
指标 Indicators	生长期 Growth period	生长旺盛期 Vigorous growth period	落叶期 Deciduous period	多时相 Multi-temporal	生长期 Growth period	生长旺盛期 Vigorous growth period	落叶期 Deciduous period	多时相 Multi-temporal
PA(%)	84.2	80.0	88.9	89.3	87.4	80.2	90.5	95.4
UA(%)	97.3	89.9	98.9	93.5	98.2	93.1	99.4	99.2
OA(%)	83.5	80.8	85.9	88.2	88.3	80.9	90.6	93.7
Kappa	0.71	0.71	0.73	0.74	0.74	0.71	0.82	0.89