落叶松人工林是我国森林资源的重要组成， 在内蒙和东北地区种植较多， 目前， 我  国人工造林工程进展迅速， 极大减轻了我国对木材供给日益增长的需求压力， 对天然林  资源起到保护作用， 而落叶松人工林是该工程的重要支柱。随着人工林项目的日益兴盛，  也促进了林业领域的发展， 其中落叶松人工林的资源调查作为林业领域的最新需求， 对  其分布信息方面的提取更是重中之重。遥感技术在落叶松人工林提取中发挥着重要作用， 为提高分类效果找到了新的方向。数据源、特征提取以及分类方法等对遥感影像分类有  重要影响，为得到更精确的分类结果， 需要基于这三个方面进行研究。

本文研究对象是落叶松人工林， 研究区选取了孟家岗林场， 以 Landsat 8 与 Sentinel-  1/2 影像数据作为数据源，利用实地调查数据、数字高程模型（Digital Elevation Model， DEM）、小斑数据资料与地面样地数据作为辅助， 结合研究区物候特性， 确定研究影像，  提取光谱特征、雷达特征和地形因子， 基于特征选择算法进行植被指数、纹理特征的提 取，构建了易于区分落叶松人工林的特征组合， 实验研究出叶绿素红边归一化指数特征、 红边归一化指数特征、近红外红边归一化指数特征等新建红边指数， 探究了新植被指数 对提取落叶松人工林精度的影响。同时， 进一步考虑到不同类别特征的最佳特征算法不 同， 构成了不同特征算法组合， 通过分类算法判别其在提取分类上的影响。本文的主要 研究内容和结论如下：

（1）多特征融合影像的落叶松人工林提取。实验对多时相 Sentinel- 1/2 影像、Landsat 8 与 Sentinel- 1 影像以及 Landsat 8 与 Sentinel- 1/2 影像进行特征提取，同时对 3 组单时相 影像提取特征，分别构建了 3 组特征集，采用 SVM 分类方法， 探讨研究出的新指数特 征对提取落叶松人工林精度的影响。实验表明加入新指数特征的单时相影像， 提取精度 明显提高， 分别提高了 3.05%、6.74%及 6.81%，且加入新指数特征的多时相影像提取精 度提高了 1.48% 、0.16% 、0.21% 。因此， 证明了新指数特征增强了分类效果， 更有利于 落叶松人工林的提取。

（2）基于特征选择算法的落叶松人工林提取。针对月份不同的遥感影像提取的特 征也不同，实验通过对各月份影像进行精度对比，确定了7 月份的影像为主要实验对象。

采用递归特征消除算法（RFE）、单变量特征选择算法、基于正则化模型算法以及基于平 均准确率的随机森林模型对单时相的 7 月份影像和多时相影像进行特征选择， 采用支持 向量机（SVM）方法分类， 通过方案对比， 发现多时相实验组的分类精度明显比单时相 影像的精度高。其中基于平均准确率的随机森林模型所选取的最佳特征组合提取精度最 高。因此， 落叶松人工林提取时多时相影像比单时相影像分类效果更好，而采用基于平 均准确率的随机森林模型也有利于落叶松人工林的提取。

（3）基于不同特征选择算法组合的落叶松人工林提取。为提高单时相影像的分类 精度，探讨研究了单时相影像植被指数与纹理特征的最佳组合， 采用递归特征消除算法、 单变量特征选择算法、基于正则化模型算法以及基于平均准确率的随机森林模型对 3 组 影像各月份的两类特征进行特征选择，采用 SVM 分类，通过精度对比进行算法组合，  选取各月份的最佳特征选择算法组合， 进行落叶松人工林提取。 研究发现，最佳特征算 法组合对分类精度提高有效。

（4）基于不同分类算法的落叶松人工林提取。实验采取了最佳特征选择算法组合， 通过对单时相影像采用 K 最近邻法（KNN）、最大似然法（MLC）、支持向量机（SVM）、 决策树算法（DTC）和随机森林法（RF）进行落叶松人工林的提取分类， 并以 1、5、7 、 10 月份影像进行实验， 研究发现，在单时相影像中支持向量机和随机森林法的精度明显 要高于其他分类算法， 而随机森林法普遍情况下要稍好于支持向量机。研究证明支持向 量机和随机森林法能够得到丰富的特征信息， 从而提高提取分类精度。因此， 基于最佳 特征算法组合的单时相影像采用支持向量机和随机森林法，分类效果明显提高。

Larch plantation is an important component of China's forest resources, and it is planted more in Inner Mongolia and Northeast China. At present, China's afforestation project is progressing rapidly, which greatly reduces the pressure of China's growing demand for wood supply and plays a role in protecting natural forest resources. Larch plantation is an important pillar of the project. With the increasing prosperity of plantation projects, it also promotes the development of forestry field. The resource survey of larch plantation is the latest demand of forestry field, and the extraction of its distribution information is the most important. Remote sensing technology plays an important role in the extraction of larch plantation and has found a new direction for improving the classification effect. Data source, feature extraction and classification methods have an important impact on remote sensing image classification. In order to obtain more accurate classification results, it is necessary to conduct research based on these three aspects.
The research object of this paper is larch plantation. Mengjiagang Forest Farm was selected in the research area. Landsat 8 and Sentinel-1/2 image data were used as data sources, field survey data, digital elevation model, small spot data and ground sample data were used as auxiliary data, and phenological characteristics of the study area were combined. Spectral features, radar features and terrain factors were extracted, index and texture features were extracted based on the feature selection algorithm, and a feature combination that was easy to distinguish larch plantations was constructed. New red edge indices such as chlorophyll normalized index, red edge normalized index and near infrared normalized index were introduced in the experiment. The effect of new vegetation index on the extraction accuracy of larch plantation was studied. At the same time, further considering the different best feature algorithms of different categories of features, the combination of different feature algorithms is formed, and the influence on extraction classification is judged by classification algorithm. The main research contents and conclusions of this paper are as follows:
（1）Extraction of larch plantation from multi-feature fusion images. In the experiment, Sentinel-1/2 image, Landsat 8 and Sentinel-1 image and Landsat 8 and Sentinel-1/2 image were fused for multi-temporal images, and features were extracted from the fused images. At the same time, three groups of feature sets were constructed respectively. SVM classification method was used to study the effect of newly established index features on the extraction accuracy of larch plantation. Experiments show that the extraction accuracy of single-phase images with new index features is improved significantly, by 3.05%, 6.74% and 6.81%, respectively, and the extraction accuracy of multi-temporal images with new index features is increased by 1.48%, 0.16% and 0.21%. Therefore, it is proved that the new index features enhance the classification effect and are more conducive to the extraction of larch plantation.
（2）Extraction of larch plantation based on feature selection algorithm. The features extracted from single-phase images in different months are also different. By comparing the accuracy of the images in each month, the image in July is determined as the main experimental object. Recursive feature elimination algorithm, single variable feature selection algorithm, regularized model algorithm and random forest model based on average accuracy were used to select the features of single-phase July image and multi-temporal fusion image, and support vector machine method was used to classify them. Through comparison of schemes, it was found that the classification accuracy of multi-temporal experimental group was significantly higher than that of single-phase image. The random forest model based on average accuracy has the highest extraction precision of the best feature combination. Therefore, the classification effect of multi-temporal image extraction is better than that of single temporal image extraction, and the random forest model based on average accuracy is also beneficial to the extraction of larch plantation.
（3）Extraction of larch plantation based on combination of different feature selection algorithms. By studying the best combination of the two types of features of the single-phase image, recursive feature elimination algorithm, univariate feature selection algorithm, regularized model algorithm and random forest model based on average accuracy were used to select the two types of features of the three groups of images in each month, and the optimal features extracted by the four feature selection algorithms for the two types of features of the single-phase image were explored. SVM classification was adopted, and the algorithm combination was carried out by precision comparison. The best feature selection algorithm combination of each month was selected to extract larch plantation. It is found that the combination of the best feature algorithm is effective to improve the classification accuracy.
（4）Extraction of larch plantation based on different classification algorithms. The experiment made use of the combination of the best feature selection algorithm, and used K nearest neighbor method, maximum likelihood method, support vector machine, decision tree algorithm and random forest method to extract and classify the single phase image. The experiment was carried out with the images in January, May, July and October. The accuracy of SVM and random forest method in single phase image is obviously higher than other classification algorithms, and random forest method is slightly better than support vector machine in general. It is proved that support vector machine and random forest method can obtain rich feature information, so as to improve the extraction classification accuracy. Therefore, the classification effect of single-phase images based on the combination of the best feature algorithm is obviously improved by using support vector machine and random forest method.

84

[1] 张颖, 李晓格, 温亚利. 碳达峰碳中和背景下中国森林碳汇潜力分析研究[J]. 北京林业大学学报, 2022, 44(01): 38-47.

[2] 张丽云, 赵天忠, 夏朝宗, 等. 遥感变化检测技术在林业中的应用[J]. 世界林业研究, 2016, 29(02): 44-48.

[3] 于德龙. 遥感技术在林业中的应用现状与展望[J]. 科技创新与应用, 2017, 18: 291.[4] 李增元, 覃先林, 高志海, 等. 高分遥感林业应用研究[J]. 卫星应用, 2018, 11: 61-65.[5] 王怀警, 谭炳香, 王晓慧, 等. 多分类器组合森林类型精细分类[J]. 遥感信息, 2019,

34(02): 104-112.

[6] 郭航, 张晓丽. 基于遥感技术的植被分类研究现状与发展趋势[J]. 世界林业研究,

2007, 3: 14-19.

[7] Fassnacht F E, Latifi H, Stere ń czak K, et al. Review of studies on tree species classification from remotely sensed data[J]. Remote Sensing of Environment, 2016, 186: 64-87.

[8] 胡杰, 张莹, 谢仕义. 国产遥感影像分类技术应用研究进展综述[J]. 计算机工程与应用, 2021, 57(03): 1-13.

[9] 颜伟, 周雯, 易利龙, 等. 森林类型遥感分类及变化监测研究进展[J]. 遥感技术与应用, 2019, 34(03): 445-454.

[10] 董心玉, 范文义, 田甜. 基于面向对象的资源 3 号遥感影像森林分类研究[J]. 浙江农林大学学报, 2016, 33(05): 816-825.

[11] 张智超, 范文义, 孙舒婷. 基于多种分类器组合的森林类型信息提取技术研究[J].森林工程, 2015, 31(03): 75-80.

[12] 王怀警, 谭炳香, 房秀凤, 等. C5.0 决策树 Hyperion 影像森林类型精细分类方法[J]. 浙江农林大学学报, 2018, 35(04): 724-734.

[13] 张晓羽, 李凤日, 甄贞, 等. 基于随机森林模型的陆地卫星 Landsat-8 遥感影像森林植被分类[J]. 东北林业大学学报, 2016, 44(06): 53-57+74.

[14] Alam S M R, Hossain M S. A rule-based classification method for mapping saltmarsh land-cover in south-eastern Bangladesh from Landsat-8 OLI[J]. Canadian Journal of Remote Sensing, 2020, 2: 1-25.

[15] 雷光斌, 李爱农, 谭剑波, 等. 基于多源多时相遥感影像的山地森林分类决策树模型研究[J]. 遥感技术与应用, 2016, 31(01): 31-41.

[16] 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

参考文献

85

[J]. 内蒙古农业大学学报:自然科学版, 2010, 2: 6.

[17] 王凯. 基于 SPOT5 森林资源分类研究[D]. 杭州: 浙江农林大学, 2014.

[18] 李军玲, 庞勇, 李增元, 等. 机载 AISA Eagle Ⅱ高光谱数据在温带天然林树种分类中的应用[J]. 东北林业大学学报, 2019, 47(5): 5.

[19] Shang X, Chisholm L A. Classification of Australian native forest species using hyperspectral remote sensing and machine-learning classification algorithms[J]. IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 2014, 7(6): 2481-2489.

[20] Fassnacht F E, Neumann C, Forster M, et al. Comparison of feature reduction algorithms for classifying tree species with hyperspectral data on three central European test sites[J]. IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 2014, 7(6): 2547-2561.

[21] 王长青, 李贝贝, 朱瑞飞, 常守志. 哨兵一号协同吉林一号影像的树种识别研究[J]. 森林工程, 2020, 36(02): 40-48.

[22] 李萌, 年雁云, 边瑞, 白艳萍, 马金辉. 基于多源遥感影像的青海云杉和祁连圆柏分类[J]. 遥感技术与应用, 2020, 35(04): 855-863.

[23] 毕恺艺, 牛铮, 黄妮, 康峻, 裴杰. 基于 Sentinel-2A 时序数据和面向对象决策树方法的植被识别[J]. 地理与地理信息科学, 2017, 33(05): 16-20+27+127.

[24] 于婉婉, 徐凯健, 赵萍, 等. Sentinel-2 影像红边谱段对不同生长期区域优势树种识别的影响[J]. 地理与地理信息科学, 2021, 37(03): 42-49.

[25] Persson M, Lindberg E, Reese H. Tree species classification with multi-temporal Sentinel-2 data[J]. Remote Sensing, 2018, 10(11): 1794.

[26] Sharifi A, Amini J, Sumantyo J T S, et al. Speckle reduction of PolSAR images in forest regions using fast ICA algorithm[J]. Journal of the Indian Society of Remote Sensing, 2015, 43(2): 339-346.

[27] Koch H B. Exploring full-waveform LiDAR parameters for tree species classification[J]. International Journal of Applied Earth Observation and Geoinformation, 2010, 13(1): 152-160.

[28] 戴鹏钦, 丁丽霞, 刘丽娟, 董落凡, 黄依婷. 基于 FCN 的无人机可见光影像树种分类[J]. 激光与光电子学进展, 2020, 57(10): 36-45.

[29] 陈向宇, 云挺, 薛联凤, 刘应安. 基于激光雷达点云数据的树种分类[J]. 激光与光电子学进展, 2019, 56(12): 203-214.

[30] 鲁续坤. 基于机载 LiDAR 和高光谱数据的树种分类及三维显示[D]. 成都: 电子科技大学, 2018, 9: 85.

西安科技大学全日制工程硕士学位论文

86

[31] 吴艳双. 基于机载高光谱和 LiDAR 数据的树种分类[D]. 北京: 北京林业大学, 2019.[32] Laurin G V, Puletti N, Hawthorne W, et al. Discrimination of tropical forest types,

dominant species, and mapping of functional guilds by hyperspectral and simulated multispectral Sentinel-2 data[J]. Remote Sensing of Environment, 2016, 176: 163-176.

[33] Na X,Zang S, Liu L, et al. Wetland mapping in the Zhalong National Natural Reserve, China, using optical and radar imagery and topographical data[J]. Journal of Applied Remote Sensing, 2013, 7(1): 609-618.

[34] Sothe C, De Almeida C M, Schimalski M B, et al. A comparison of machine and deeplearning algorithms applied to multisource data for a subtropical forest area classification[J]. International Journal of Remote Sensing, 2020, 41(5): 1943-1969.

[35] 李明泽, 付瑜, 于颖, 等. 基于多时相 SAR 数据和 SPOT 数据的盘古林场林分类型识别[J]. 植物研究, 2016, 36(04): 613-619+626.

[36] Gill T K, Phinn S R, Armston J D, et al. Estimating tree-cover change in Australia: challenges of using the MODIS vegetation index product[J]. International Journal of Remote Sensing, 2009, 30(6): 1547-1565.

[37] 刘峰, 张贵. 基于 GIS 和 RS 的广州市森林植被分类研究[J]. 湖南林业科技, 2004, 1:

15-17.

[38] Nemani R, Running S. Land Cover Characterization Using Multitemporal Red, Near-IR, and Thermal-IR Data from NOAA/AVHRR[J]. Ecological Applications, 1997, 7(01): 79.

[39] Mcdonald A J, Gemmell F M, Lewis P E. Investigation of the Utility of Spectral Vegetation Indices for Determining Information on Coniferous Forests[J]. Remote Sensing of Environment, 1998, 66(03): 250-272.

[40] Key T, Warner T A, Mcgraw J B, et al. A Comparison of Multispectral and Multitemporal Information in High Spatial Resolution Imagery for Classification of Individual Tree Species in a Temperate Hardwood Forest[J]. Remote Sensing of Environment, 2001, 75(1): 100-112.

[41] 窦刚, 陈广胜, 赵鹏. 采用颜色纹理及光谱特征的木材树种分类识别[J]. 天津大学学报(自然科学与工程技术版), 2015, 48(02): 147-154.

[42] 岳俊, 王振锡, 冯振峰, 李子艺, 王玲段. 基于光谱与纹理特征的南疆盆地果树树种遥感识别研究[J]. 新疆农业大学学报, 2015, 38(04): 326-333.

[43] Le W, Sousa W P, Peng G, et al. Comparison of IKONOS and QuickBird images for mapping mangrove species on the Caribbean coast of Panama[J]. Remote Sensing of Environment, 2004, 91(3-4): 432-440.

[44] 陈工, 李琦, 张彦南, 等. 多源遥感信息提取桉树人工林[J]. 浙江林业科技, 2018,

参考文献

87

38(02): 78-87.

[45] 何云, 黄翀, 李贺, 刘庆生, 刘高焕, 周振超, 张晨晨. 基于 Sentinel-2A 影像特征优选的随机森林土地覆盖分类[J]. 资源科学, 2019, 41(05): 992-1001.

[46] 王娜, 李强子, 杜鑫, 等. 单变量特征选择的苏北地区主要农作物遥感识别[J]. 遥感学报, 2017, 21(04): 519-530.

[47] 刘家福, 李林峰, 任春颖, 等. 基于特征优选的随机森林模型的黄河口滨海湿地信息提取研究[J]. 湿地科学, 2018, 16(02): 97-105.

[48] Cheng K, Wang J. Forest type classification based on integrated spectral-spatial-temporal features and random forest Algorithm-A case study in the Qinling Mountains[J]. Forests, 2019, 10(7): 559.

[49] 周小成, 郑磊, 黄洪宇. 基于多特征优选的无人机可见光遥感林分类型分类[J]. 林业科学, 2021, 57(06): 24-36.

[50] Maxwell A E, Warner T A, Fang F. Implementation of machine-learning classification in remote sensing: An applied review[J]. International Journal of Remote Sensing, 2018, 39(9): 2784-2817.

[51] 林大辉, 陈秋妹, 宁正元. 基于支持向量机的栗属树种分类研究[J]. 莆田学院学报,

2009, 16(05): 39-42+46.

[52] 刘怀鹏, 安慧君. 利用最大似然法识别呼和浩特市绿化树种[J]. 东北林业大学学报,

2014, 42(07): 157-160+169.

[53] Hagner O, Reese H. A method for calibrated maximum likelihood classification of forest types[J]. Remote Sensing of Environment, 2007, 110(04): 438-444.

[54] Michele D, Liviu T E, Mattia M, Terje G, Erik N. Semi-supervised SVM for individual tree crown species classification[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 110: 77-87.

[55] 邓曾, 李丹, 柯樱海, 吴燕晨, 李小娟, 宫辉力. 基于改进 SVM 算法的高分辨率遥感影像分类[J]. 国土资源遥感, 2016, 28(03): 12-18.

[56] Haapanen R, Ek A R, Bauer M E, et al. Delineation of forest/non-forest land use classes using nearest neighbor methods[J]. Remote Sensing of Environment, 2004, 89(3): 265-271.

[57] 陈丽萍, 孙玉军. 基于不同决策树的面向对象林区遥感影像分类比较[J]. 应用生态学报, 2018, 29(12): 3995-4003.

[58] Berberoglu S, Yilmaz K T, Özkan C. Mapping and monitoring of coastal wetlands of Cukurova Delta in the Eastern Mediterranean region[J]. Biodiversity & Conservation, 2004, 13(3): 615-633.

西安科技大学全日制工程硕士学位论文

[59] Furuya D E G, Aguiar J A F, Estrabis N V, et al. A Machine Learning Approach for Mapping Forest Vegetation in Riparian Zones in an Atlantic Biome Environment Using Sentinel-2 Imagery[J]. Remote Sensing, 2020, 12(24): 4086.

[60] Tian X, Chen L, Zhang X. Classifying tree species in the plantations of southern China based on wavelet analysis and mathematical morphology[J]. Computers & Geosciences, 2021, 151: 104757.

[61] 张智超, 范文义, 孙舒婷. 基于多种分类器组合的森林类型信息提取技术研究[J]. 森林工程, 2015, 31(03): 75-80.

[62] 王怀警. 森林类型高光谱遥感分类研究[D]. 北京: 中国林业科学研究院, 2018.[63] 郭瑞霞. 基于多源数据的落叶松人工林识别研究[D]. 西安: 西安科技大学, 2019.

[64] 韩婷婷, 习晓环, 王成, 等. 基于决策树方法的云南省森林分类研究[J]. 遥感技术与应用, 2014, 29(05): 744-751.

[65] 李若楠, 欧光龙, 代沁伶, 等. 基于 GEE 和 Landsat 时间序列数据的香格里拉森林类型分类研究[J]. 西南林业大学学报(自然科学), 2020, 40(05): 115-125.

[66] Schiefer F, Kattenborn T, Frick A, et al. Mapping forest tree species in high resolution UAV-based RGB-imagery by means of convolutional neural networks[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2020, 170: 205-215.

[67] Mäyrä J, Keski-Saari S, Kivinen S, Hurskainen P, Poikolainen L, Viinikka A, Tuominen S, Kumpula T, Vihervaara P. Tree species classification from airborne hyperspectral and LiDAR data using 3D convolutional neural networks[J]. Remote Sensing of Environment, 2021, 256: 112322.

[68] Pearse G D, Watt M S, Soewarto J, et al. Deep learning and phenology enhance large-scale tree species classification in aerial imagery during a biosecurity response[J]. Remote Sensing, 2021, 13(9): 1789.

[69] Flood N, Watson F, Collett L. Using a U-net convolutional neural network to map woody vegetation extent from high resolution satellite imagery across Queensland, Australia[J]. International Journal of Applied Earth Observation and Geoinformation ,2019, 82: 101897.

[70] 滕文秀, 温小荣, 王妮, 等. 基于深度迁移学习的无人机高分影像树种分类与制图[J]. 激光与光电子学进展, 2019, 56(07): 277-286.

[71] 王雅慧, 陈尔学, 郭颖, 等. 高分辨率多光谱遥感影像森林类型分类深度 U-net 优化方法[J]. 林业科学研究, 2020, 33(01): 11-18.