欢迎访问作物学报，今天是
作物学报

作物学报 ›› 2018, Vol. 44 ›› Issue (04): 569-580.doi: 10.3724/SP.J.1006.2018.00569

• 耕作栽培·生理生化 • 上一篇    下一篇

随机森林方法在玉米-大豆精细识别中的应用

王利民(), 刘佳, 杨玲波, 杨福刚, 富长虹   

  1. 中国农业科学院农业资源与农业区划研究所, 北京 100081
  • 收稿日期:2017-05-08 接受日期:2018-01-08 出版日期:2018-01-26 网络出版日期:2018-01-26
  • 作者简介:

    wanglimin01@caas.cn

  • 基金资助:
    本研究由国家“十二五”科技重大专项(09-Y30B03-9001-13/15)和国家重点研发计划项目(2016YFD0300603)资助

Application of Random Forest Method in Maize-soybean Accurate Identification

Li-Min WANG(), Jia LIU, Ling-Bo YANG, Fu-Gang YANG, Chang-Hong FU   

  1. Institute of Agricultural Resources and Regional Planning, Chinese Academy of Agricultural Sciences, Beijing 100081, China
  • Received:2017-05-08 Accepted:2018-01-08 Published:2018-01-26 Published online:2018-01-26
  • Supported by:
    This study was supported by the National Science and Technology Major Project (09-Y30B03-9001-13/15) and the National Key Research and Development Program of China (2016YFD0300603).

RichHTML

21

PDF

1180

被引次数

1

摘要：

研究基于遥感影像的作物精确识别技术方法, 对获取作物分布信息具有重要意义。随机森林分类(random forest classification, RFC)是机器学习的一种, 本文使用Landsat-8 OLI卫星影像数据, 针对研究区内的大豆、玉米和其他地物等3种主要作物类型, 系统比较了该方法与较为成熟的最大似然分类(maximum likelihood classification, MLC)、支持向量机分类(support vector machine, SVM)方法的分类精度。结果表明, MLC、SVM、RFC的总体分类精度分别为91.68%、91.49%、94.32%, Kappa系数分别为0.87、0.87、0.91, RFC方法作物识别精度比MLC和SVM分类显著提升。对原始7波段影像进行主成分变换(principal component analysis, PCA), 提取前4个主成分分量, 同时计算归一化植被指数(normalized difference vegetation index, NDVI)和归一化水体指数(normalized difference water index, NDWI), 将6个额外辅助特征波段叠加到原始7个波段影像上进行再次分类, MLC和SVM方法作物识别精度未有提升, RFC方法总体精度提高了1.49个百分点, Kappa系数提高0.03, 精度提升幅度有限, 主要原因是6个辅助波段在类型识别中作用较小。在分类耗时上, MLC、SVM、RFC分别为145 s、11 000 s、1800 s, 表明随机森林分类具有最好的分类精度和适中的耗时。综合评价后, 随机森林分类方法在进行大豆-玉米精细识别中具有较大优势, 具有业务应用的潜力。

关键词: Landsat-8, 随机森林, 玉米, 大豆, 遥感, 识别能力

Abstract:

It is very important to obtain the crop identification information based on remote sensing image. Remote sensing images have the advantages of high efficiency, high accuracy, low costs, and wide monitoring scope. Applying remote sensing images in maize-soybean accurate identification and planting area evaluation can give full play to the advantages of remote sensing images. Random forest classification (RFC) is a new classification method, a type of machine learning. Currently, there are very few studies on crop classification based on RFC. In order to evaluate the potential of the method on maize-soybean crop accurate identification, the paper conducted classification of major crops of soybean, maize, and other ground objects. Utilizing Landsat-8 OLI satellite image data, and three methods including maximum likelihood classification (MLC), support vector machine (SVM), and random forest classification (RFC). The overall classification accuracies of MLC, SVM, and RFC were 91.68%, 91.49%, and 94.32%, with their kappa coefficients of 0.87, 0.87, and 0.91, respectively, showing that RFC is better. The principal component analysis (PCA) was made on original seven wave band images, and the first four wave bands of the major components were extracted. Meanwhile, the normalized difference vegetation index (NDVI) and normalized difference water index (NDWI) were calculated; six additional supporting characteristic wave bands were overlapped on original seven wave band images, and the classifications with MLC, SVM, and RFC were conducted again. After adding characteristic wave bands, crop identification accuracies by MLC and SVM methods were not improved. The accuracy of RFC method was increased slightly with overall accuracy of 95.81% increasing by 1.49 percent, and Kappa coefficient of 0.94 increasing by 0.03, showing accuracy slightly increased, and limited improvement effect. Near-infrared band and two short infrared wave bands were most important, while newly added wave band was not significant for soybean-maize identification, showing the limited improvement effect of supporting wave band. SVM had the longest time spent on classification, with about 11 000 s; MLC the least, only 145 s; and RFC about 1800 s. It indicates that SVM doesn’t have any advantages in both accuracy and time-consumed, however, MLC can quickly get the classification results, and RFC has the highest classification accuracy with moderate time consumed. In conclusion, RFC has greater advantage in soybean-maize accurate identification, and is suitable to be widely applied in the operation of regional agriculture remote sensing monitoring crop area extraction.

Key words: Landsat-8, random forest, maize, soybean, remote sensing, identification capacity

图1

研究区地理位置"

表1

Landsat 8 OLI影像各波段辐射定标系数"

波段
Band		 定标斜率
Gain		 定标截距
Bias
海岸蓝波段 Coastal aerosol 0.01298 -64.89967
蓝 Blue 0.013292 -66.45805
绿 Green 0.012248 -61.24053
红 Red 0.010328 -51.64146
近红外 Near infrared 0.0063204 -31.602
短波红外1 SWIR 1 0.0015718 -7.85913
短波红外2 SWIR 2 0.00052979 -2.64895

图2

研究区Landsat 8 OLI影像及样方分布 a: Landsat 8原始影像及样方分布; b: 样方原图; c: 样方分类结果。"

图3

研究区主要地物类别光谱曲线"

图4

研究技术流程"

图5

基于RapidEye影像的目视解译结果 a: Rapideye影像(5/4/3波段); b: Rapideye影像的目视解译结果。"

图6

基于原始影像3种方法分类结果 a: 最大似然分类结果; b: 支持向量机分类结果; c: 随机森林分类结果; d: 最大似然分类结果局部; e: 支持向量机分类结果局部; f: 随机森林分类结果局部。"

表2

基于原始影像3种分类方法的混淆矩阵"

作物
Crop		 分类方法
Method		 大豆
Soybean (pixel)		 玉米
Maize (pixel)		 其他
Other (pixel)		 总计
Total (pixel)		 制图精度
Mapping accuracy (%)
大豆
Soybean		 MLC 864849 15393 56545 936787 91.03
SVM 878082 7498 66202 951782 92.42
RFC 911841 9285 65006 986132 95.98
玉米
Maize		 MLC 21443 1134475 76472 1232390 92.38
SVM 35185 1140200 93000 1268385 92.85
RFC 4158 1186858 68423 1259439 96.65
其他
Other		 MLC 63758 78172 1436164 1578094 91.52
SVM 36783 80342 1409979 1527104 89.85
RFC 34051 31897 1435752 1501700 91.50
总计 Total (pixel) 950050 1228040 1569181 3747271
用户精度
User accuracy (%)		 MLC 92.32 92.05 91.01
SVM 92.26 89.89 92.33
RFC 92.47 94.24 95.61
总体精度
Overall accuracy (%)		 MLC 91.68 Kappa系数
Kappa coefficient		 MLC 0.87
SVM 91.49 SVM 0.87
RFC 94.32 RFC 0.91

图7

增加辅助特征后的3种方法分类结果 a: 最大似然分类结果; b: 支持向量机分类结果; c: 随机森林分类结果; d: 最大似然分类结果局部; e: 支持向量机分类结果局部; f: 随机森林分类结果局部。"

表3

增加辅助特征波段后的3种分类方法分类结果混淆矩阵"

作物
Crop		 分类方法
Method		 大豆
Soybean (pixel)		 玉米
Maize (pixel)		 其他
Other (pixel)		 总计
Total (pixel)		 制图精度
Mapping accuracy (%)
大豆
Soybean		 MLC 857118 3824 56780 917722 90.22
SVM 885153 7611 66296 959060 93.17
RFC 934124 2276 55865 992265 98.32
玉米
Maize		 MLC 56343 1201206 167854 1425403 97.81
SVM 31759 1145348 87264 1264371 93.27
RFC 2053 1205962 63085 1271100 98.20
其他
Other		 MLC 36589 23010 1344547 1404146 85.68
SVM 33138 75081 1415621 1523840 90.21
RFC 13873 19802 1450231 1483906 92.42
总计 Total (pixel) 950050 1228040 1569181 3747271
用户精度
User accuracy (%)		 MLC 93.40 84.27 95.76
SVM 92.29 90.59 92.90
RFC 94.14 94.88 97.73
总体精度
Overall accuracy (%)		 MLC 90.81 Kappa系数
Kappa coefficient		 MLC 0.86
SVM 91.96 SVM 0.88
RFC 95.81 RFC 0.94

图8

增加辅助特征前后各特征变量的重要性 a: 增加辅助特征前各波段重要性; b: 增加辅助特征后各波段重要性。"

表4

3种分类方式作物分类提取耗费时间"

分类方式
Classification method		 分类时间
Time cost (s)
最大似然分类MLC 145
支持向量机SVM 11000
随机森林分类RFC 1800
[1] 许文波, 田亦陈. 作物种植面积遥感提取方法的研究进展. 云南农业大学学报, 2005, 20(1): 94-98
Xu W B, Tian Y C.Overview of extraction of crop area from remote sensing.J Yunnan Agric Univ, 2005, 20(1): 94-98 (in Chinese with English abstract)
[2] 尤淑撑, 孙毅, 李小文. 成像光谱技术在土地利用动态遥感监测中的应用研究. 遥感信息, 2005, (3): 31-33
You S C, Sun Y, Li X W.Reseach on landuse dynamic monitoring using high spectral resolution remote sensing data.Remote Sens Inf, 2005, (3): 31-33 (in Chinese with English abstract)
[3] Gleriani J M, da Silva J D S, Epiphanio J C N. Comparative performance of neural networks and maximum likelihood for supervised classification of agricultural crops: single date and temporal analysis.Radal Ba Fnon, 2004, 4: 2959-2964
[4] Liang Y J, Xu Z M.Crop identification in the irrigation district based on SPOT-5 satellite imagery.Pratacult Sci, 2013, 30: 161-167
[5] Baup F, Flanquart S, Maraissicre C, Fieuzal R.Satellite monitoring at high spatial resolution of water bodies used for irrigation purposes.Sci Technol Innovation Herald, 2012, 32(3): 103-119
[6] Luo B, Yang C, Chanussot J, Zhang L.Crop yield estimation based on unsupervised linear unmixing of multidate hyperspectral imagery.IEEE Trans Geosci Remote Sens, 2013, 51: 162-173
[7] Wu B, Li Q.Crop planting and type proportion method for crop acreage estimation of complex agricultural landscapes.Int J Appl Earth Obs Geoinf, 2012, 16: 101-112
[8] Long J A, Lawrence R L, Greenwood M C, Marshall L, Miller P R.Object-oriented crop classification using multitemporal ETM+ SLC-off imagery and random forest.Gisci Remote Sens, 2013, 50: 418-436
[9] Jiao X F, Kovacs J M, Shang J L, McNairn H, Walters D, Ma B L, Geng X Y. Object-oriented crop mapping and monitoring using multi-temporal polarimetric RADARSAT-2 data.ISPRS J Photogramm Remote Sens, 2014, 96: 38-46
[10] Rosales H S, Bruno C, Balzarini M.Identifying yield and environment relationships using classification and regression trees (CART). Interciencia, 2010, 35: 876-882
[11] Arvor D, Jonathan M, Simoes M, Durieux L.Classification of MODIS EVI time series for crop mapping in the state of Mato Grosso, Brazil. Int J Remote Sens, 2011, 32: 7847-7871
[12] 李鑫川, 徐新刚, 王纪华, 武洪峰, 金秀良, 李存军, 鲍艳松. 基于时间序列环境卫星影像的作物分类识别. 农业工程学报, 2013, 29(2): 169-176.
Li X C, Xu X G, Wang J H, Wu H F, Jin X L, Li C J, Bao Y S.Crop classification recognition based on time-series images from HJ satellite.Trans CSAE, 2013, 29(2): 169-176 (in Chinese with English abstract)
[13] Kaur P, Singh S, Garg S, Harmanpreet. Analytical and CASE study on limited search, ID3, CHAID, C4.5, improved C4.5 and OVA decision tree algorithms to design decision support system.Strategic Change, 2010, 1324: 253-267
[14] Deng X, Zhao C, Yan H.Systematic modeling of impacts of land use and land cover changes on regional climate: a review.Adv Meteorol, 2013, 2013: 317678
[15] 刘建光, 李红, 孙丹峰, 张微微, 周连第. MODIS土地利用/覆被多时相多光谱决策树分类. 农业工程学报, 2010, 26(10): 312-318
Liu J G, Li H, Sun D F, Zhang W W, Zhou L D.Land use/cover decision tree classification fusing multi-temporal and multi-spectral of MODIS.Trans CSAE, 2010, 26(10): 312-331 (in Chinese with English abstract)
[16] 刘毅, 杜培军, 郑辉, 夏俊士, 柳思聪. 基于随机森林的国产小卫星遥感影像分类研究. 测绘科学, 2012, 37(04): 194-196
Liu Y, Du P J, Zheng H, Xia J S, Liu S C.Classification of China small satellite remote sensing image based on random forests.Sci Surv Mapping, 2012, 37(04): 194-196 (in Chinese with English abstract)
[17] 刘磊, 江东, 徐敏, 尹芳. 基于多光谱影像和专家决策法的作物分类研究. 安徽农业科学, 2011, 39(25): 1703-1706
Liu L, Jiang D, Xu M, Yin F.Crops classification based on multi-spectral image and decision tree method. J Anhui Agric Sci, 2011, 39(25): 1703-1706 (in Chinese with English abstract)
[18] 康峻, 侯学会, 牛铮, 高帅, 贾坤. 基于拟合物候参数的植被遥感决策树分类. 农业工程学报, 2014, 30(9): 148-156
Kang J, Hou X H, Niu Z, Gao S, Jia K.Decision tree classification based on fitted phenology parameters from remotely sensed vegetation data.Trans CSAE, 2014, 30(9): 148-156 (in Chinese with English abstract)
[19] 张旭东, 迟道才. 基于异源多时相遥感数据决策树的作物种植面积提取研究. 沈阳农业大学学报, 2014, 45: 451-456
Zhang X D, Chi D C.Mapping crop fields by using multi-sensor and multi-temporal remote sensing data with decision-tree.J Shenyang Univ, 2014, 45: 451-456 (in Chinese with English abstract)
[20] 黄健熙, 贾世灵, 武洪峰, 苏伟. 基于GF-1 WFV影像的作物面积提取方法研究. 农业机械学报, 2015, 46(1): 253-259
Huang J X, Jia S L, Wu H F, Su W.Extraction method of crop planted area based on GF-1 WFV Image.Trans CSAM, 2015, 46(1): 253-259 (in Chinese with English abstract)
[21] Kandrika S, Roy P S.Land use land cover classification of Orissa using multi-temporal IRS-P6 awifs data: A decision tree approach.Int J Appl Earth Obs Geoinf, 2008, 10: 186-193
[22] Peña J M, Gutiérrez P A, Hervás-Martínez C, Six J, Plant R E, López-Granados F.Object-based image classification of summer crops with machine learning methods.Remote Sens, 2014, 6: 5019-5041
[23] Pal M.Random forest classifier for remote sensing classification.Int J Remote Sens, 2007, 26: 217-222
[24] Gislason P O, Benediktsson J A, Sveinsson J R.Random forests for land cover classification.Pattern Recognit Lett, 2003, 27: 294-300
[25] Ok A O, Akar O, Gungor O.Evaluation of random forest method for agricultural crop classification.Eur J Remote Sens, 2012, 45: 421-432
[26] Deschamps B, Mcnairn H, Shang J, Jiao X.Towards operational radar-only crop type classification: comparison of a traditional decision tree with a random forest classifier.Can J Remote Sens, 2012, 38: 60-68
[27] 张晓羽, 李凤日, 甄贞, 赵颖慧. 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
Zhang X Y, Li F R, Zhen Z, Zhao Y H.Forest vegetation classification of Landsat8 remote sensing image based on random forest model.J Northeast For Univ, 2016, 44(6): 53-57 (in Chinese with English abstract)
[28] 郭玉宝, 池天河, 彭玲, 刘吉磊, 杨丽娜. 利用随机森林的高分一号遥感数据进行城市用地分类. 测绘通报, 2016, (5): 73-76
Guo Y B, Chi T H, Peng L, Liu J L, Yang L N.Classification of GF-1 remote sensing image based on random forests for urban land-use.Bull Surv Mapping, 2016, (5): 73-76 (in Chinese with English abstract)
[29] 黄健熙, 侯矞焯, 苏伟, 刘峻明, 朱德海. 基于GF-1 WFV数据的玉米与大豆种植面积提取方法. 农业工程学报, 2017, 33(7): 164-170
Huang J X, Hou Y Z, Su W, Liu J M, Zhu D H.Mapping corn and soybean cropped area with GF-1 WFV data.Trans CSAE, 2017, 33(7): 164-170 (in Chinese with English abstract)
[30] 王利民, 刘佳, 杨玲波, 杨福刚, 富长虹. 短波红外波段对玉米大豆种植面积识别精度的影响. 农业工程学报, 2016, 32(19): 169-178
Wang L M, Liu J, Yang L B, Yang F G, Fu C H.Impact of short infrared wave band on identification accuracy of corn and soybean area.Trans CSAE, 2016, 32(19): 169-178 (in Chinese with English abstract)
[31] 王增林, 朱大明. 基于遥感影像的最大似然分类算法的探讨. 河南科学, 2010, 28: 1458-1461
Wang Z L, Zhu D M.A study of maximum likelihood classification algorithm based on remote sensing image.Henan Sci, 2010, 28: 1458-1461 (in Chinese with English abstract)
[32] Cortes C, Vapnik V.Support-vector networks.Mach Learn, 1995, 20: 273-297
[33] Breiman L.Random forests.Machine Learning, 2001, 45: 5-32
[34] Congalton R G.A Review of assessing the accuracy of classifications of remotely sensed data.Remote Sens Environ, 1991, 37: 35-46
[35] Hay A M.The derivation of global estimation from a confusion matrix.Int J Remote Sens, 1988, 9: 1395-1398
[36] Congalton R G.A comparison of sampling schemes used in generating error matrices for assessing the accuracy of maps generated from remotely sensing data. Photogramm Eng Remote Sens, 1988, 54: 593-600
[1] 刘恩波, 陈静, 李红星, 于宁宁, 任佰朝, 赵斌, 刘鹏, 张吉旺. 遮阴改变源-库平衡和调节碳水化合物代谢进而抑制夏玉米幼穗发育[J]. 作物学报, 2026, 52(6): 1891-1901.
[2] 梁进宇, 尹嘉德, 王红丽, 张国平, 侯慧芝, 董博, 马明生. 基于无人机高光谱和机器学习的旱地饲用玉米叶片氮含量估测[J]. 作物学报, 2026, 52(6): 1788-1801.
[3] 金昱何, 王雪菲, 徐张一娃, 缪怡宁, 蒋云杰, 伊莹, 缪德麟, 朱静仪, 钟一帆, 陈铭亨, 方芳, 刘鹏. 外源激素对低温胁迫下大豆叶片叶绿素荧光参数及抗氧化酶系统的影响[J]. 作物学报, 2026, 52(6): 1817-1829.
[4] 唐宽强, 李公允, 宋美毅, 赵雪, 常春玲. 大豆株高性状全基因组关联分析及预测模型构建[J]. 作物学报, 2026, 52(6): 1743-1756.
[5] 孙淑凤, 许振南, 黄嘉鑫, 翁建峰, 李新海. 玉米MAPK家族全基因组鉴定及其对拟轮枝镰孢菌感染的响应[J]. 作物学报, 2026, 52(5): 1291-1308.
[6] 姚术, 郭凯悦, 翟慧慧, 姚佳慧, 邓文琪, 闫玲, 黄驰, 高阳, 俞嫣然, 赵振邦, 李英慧, 王晓波, 李佳佳. 大豆苗期耐低铁综合评价及优异种质筛选[J]. 作物学报, 2026, 52(5): 1373-1387.
[7] 张宁宁, 滕雨菲, 任娜娜, 魏兴卓, 闫书豪, 樊可心, 王永宏, 陈文康, 张兴华, 朱万超, 徐淑兔, 薛吉全. 201份玉米自交系抗旱表型评价及可塑性分析[J]. 作物学报, 2026, 52(5): 1309-1325.
[8] 杨扬, 常诗惠, 田红丽, 易红梅, 王璐, 任洁, 范亚明, 刘亚维, 王凤格, 赵久然. 不同生态区国审玉米品种的遗传多样性分析[J]. 作物学报, 2026, 52(5): 1352-1364.
[9] 张鸿蓉, 王菲儿, 李盼, 仇海龙, 朱静, 赵连豪, 南运有, 何蔚, 樊志龙, 胡发龙, 柴强, 殷文. 减量20%灌水与25%有机肥替代化肥提高青贮玉米产量的光合特性[J]. 作物学报, 2026, 52(5): 1487-1500.
[10] 杨欣雨, 崔文涛, 迪力尼格尔·阿力木, 汪凯翔, 吴鹏昊, 任姣姣. 玉米穗上叶片数全基因组关联分析和全基因组选择[J]. 作物学报, 2026, 52(5): 1573-1590.
[11] 韩亚鑫, 何冠华, 张小琼, 张登峰, 李永祥, 刘旭洋, 王天宇, 黎裕, 邹华文, 李春辉. 基于RNA-Seq和BSA-Seq联合分析挖掘玉米侧根密度基因资源[J]. 作物学报, 2026, 52(5): 1341-1352.
[12] 蔡宏玮, 于爱忠, 姜科强, 王鹏飞, 王玉珑, 霍建喆, 庞小能, 尹波, 尚永盼. 干旱灌区有机肥替代部分化肥促进甜玉米产量提升的关键机制[J]. 作物学报, 2026, 52(4): 1166-1180.
[13] 田红丽, 杨扬, 范亚明, 易红梅, 郭丹丹, 王凤格, 赵久然. 适于玉米品种鉴定的一套三等位变异SNP新型标记组合[J]. 作物学报, 2026, 52(4): 993-1005.
[14] 杨亚莉, 徐明睿, 马越飞, 海艺蕊, 刘凯栋, 刘万茂, 孙颖. 玉米根尖及整根响应缺铁的转录组比较研究[J]. 作物学报, 2026, 52(4): 1006-1021.
[15] 张超, 郭欢, 李忠玲, 岳淑宁, 赵娜. 基于BSA-seq技术定位玉米籽粒花青素关联基因[J]. 作物学报, 2026, 52(3): 780-789.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
京ICP备15008789号-2