基于随机森林特征选择的茶园遥感提取

doi:10.13229/j.cnki.jdxbgxb20210138

摘要/Abstract

摘要：

由于茶园空间分布零散、形状不规则、与周围植被光谱特征接近等原因，从卫星影像中提取茶园非常具有挑战性。针对这一问题，本文研究提出了基于随机森林特征选择方法和Landsat-8 OLI影像进行茶园提取的技术路线，以浙江省安吉县为例，采用春、秋、冬季Landsat-8 OLI影像作为主要数据源，利用随机森林对茶园影像特征进行重要性评估、排序和特征选择，设计了单季节初始特征集、单季节优选特征集、多季节优选特征集，并进行了9组茶园提取实验。结果表明，融合了特征选择和多季节信息优势于一体的多季节优选特征集具有最好的性能表现，其精度如下：生产者精度为87.5%、总体精度为92.4%、Kappa系数为0.897。本文提出的技术路线充分发挥了其在提高遥感分类精度和降低维度等方面的作用，实现了对空间分布离散、形状欠规则的茶园的有效提取。

关键词: 农业工程, 茶园, 随机森林, 特征选择, 遥感, 提取

Abstract:

Due to the scattered spatial distribution， irregular shape and close to the spectral characteristics of surrounding vegetation， it is a very challenging work to extract tea plantations from satellite images. In order to solve this problem， a technical route of tea plantations extraction based on Random Forest feature selection method and Landsat-8 OLI image was proposed. In this study， Anji County in Zhejiang Province was selected as the research area. The spring， autumn and winter landsat-8 OLI images were used as the main data source. The importance evaluation， ranking and feature selection of initial features were carried out using Random Forest. The single season initial feature set， single season optimal feature set and multi-seasonal optimal feature set were designed， and nine groups of tea garden extraction experiments were carried out. The results show that multi-seasonal optimal feature set， which combines multi-seasonal information and feature selection advantages， has the best performance （PA = 87.5%； OA = 92.4%； Kappa = 0.897）. The technical route is reliable and practical for extracting tea plantations with scattered spatial distribution and irregular shape， and achieves relatively high accuracy while feature dimensionality reduces.

Key words: agricultural engineering, tea plantations, random forest, feature selection, remote sensing, extraction

中图分类号:

F307.1

王斌,何丙辉,林娜,王伟,李天阳. 基于随机森林特征选择的茶园遥感提取[J]. 吉林大学学报(工学版), 2022, 52(7): 1719-1732.

Bin WANG,Bing-hui HE,Na LIN,Wei WANG,Tian-yang LI. Tea plantation remote sensing extraction based on random forest feature selection[J]. Journal of Jilin University(Engineering and Technology Edition), 2022, 52(7): 1719-1732.

图/表 17

图1

图2

表1

图3

图4

图5

图6

表2

纹理指标"

纹理类别	纹理指标	公式	指标描述	说明
GLCM	均值	$M e a n = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) × i$	纹理规则程度	i，j是像元在图像中的行列坐标， p （i，j）是灰度联合概率矩阵，quant_k 是灰度共生矩阵的阶数。
	方差	$V a r i a n c e = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) × (i - M e a n) 2$	像元灰度值与均值的偏差量
	对比度	$C o n t r a s t = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) × (i - j) 2$	纹理清晰度
	同质性	$H o m o g e n e i t y = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) × 1 1 + (i + j) 2$	纹理均匀性
	非相似性	$D i s s m i l a r i t y = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) × i - j$	纹理对比度
	角二阶矩	$A S M = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) 2$	灰度分布均匀性
	熵	$E n t r o p y = - ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k p (i, j) × l n p (i, j)$	纹理复杂度统计量
	相关性	$C o r r e l a t i o n = ∑ i = 0 q u a n t k ∑ j = 0 q u a n t k (i - M e a n) × (j - M e a n) × p (i, j) 2 V a r i a n c e$	纹理灰度线性关系和方向性统计量
地统计学	变差函数	$g k (h) = 1 2 n (h) ∑ i = 1 n (h) {d n k (x i) - d n k (x i + h)} 2$	图像局部方差和相关性统计量	dn是像元x的灰度值， n（h）是相距为h的像元对的数量，k是波段序号。
地统计学	伪交叉变差函数	$γ j k (h) = 1 2 n (h) ∑ i = 1 n (h) {d n j (x i) - d n k (x i + h)} 2$	两个波段之间的联合变差函数	dn是像元x的灰度值， n（h）是相距为h的像元对的数量，k是波段序号。

表2

图7

表3

表4

图8

表5

图9

图10

图11

表6

参考文献 23

1	Liu Y L, Feng Y H, Zhao Z, et al. Socioeconomic drivers of forest loss and fragmentation: a comparison between different land use planning schemes and policy implications[J]. Land Use Policy, 2016, 54: 58-68.
2	Su S L, Zhou X C, Wan C, et al. Land use changes to cash crop plantations: crop types, multilevel determinants and policy implications[J]. Land Use Policy, 2016, 50: 379-389.
3	徐乔,张霄,余绍淮,等. 综合多特征的极化SAR图像随机森林分类算法[J]. 遥感学报, 2019, 23(4): 685-694.
	Xu Qiao, Zhang Xiao, Yu Shao-huai, et al. Multi-feature-based classification method using random forest and superpixels for polarimetric SAR images[J]. Journal of Remote Sensing, 2019, 23(4): 685-694.
4	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(4): 519-530.
	Wang Na, Li Qiang-zi, Du Xin, et al. Identification of main crops based on the univariate feature selection in Subei[J]. Journal of Remote Sensing, 2017, 21(4): 519-530.
5	耿仁方,付波霖,蔡江涛,等. 基于无人机影像和面向对象随机森林算法的岩溶湿地植被识别方法研究[J]. 地球信息科学学报, 2019, 21(8): 1295-1306.
	Geng Ren-fang, Fu Bo-lin, Cai Jiang-tao, et al. Object-based karst wetland vegetation classification method using unmanned aerial vehicle images and random forest algorithm[J]. Journal of Geo-Information Science, 2019, 21(8): 1295-1306.
6	董强,刘晶红,周前非. 用于遥感图像拼接的改进SURF算法[J]. 吉林大学学报: 工学版, 2017, 47(5): 1644-1652.
	Dong Qiang, Liu Jin-hong, Zhou Qian-fei. Improved SURF algorithm used in image mosaic[J]. Journal of Jilin University (Engineering and Technology Edition), 2017, 47(5): 1644-1652.
7	刘凯,龚辉,曹晶晶,等. 基于多类型无人机数据的红树林遥感分类对比[J]. 热带地理, 2019, 39(4): 492-501.
	Liu Kai, Gong Hui, Cao Jing-jing, et al. Comparison of mangrove remote sensing classification based on multi-type UAV data[J]. Tropical Geography, 2019, 39(4): 492-501.
8	陈媛媛,郑加柱,魏浩翰,等. 基于不同特征的随机森林极化SAR图像分类[J]. 计算机系统应用, 2019, 28(8): 183-189.
	Chen Yuan-yuan, Zheng Jia-zhu, Wei Hao-han, et al. Tidal flat classification based on random forest model using different features of polarimetric SAR[J]. Computer Systems & Applications, 2019, 28(8): 183-189.
9	许章华,黄旭影,林璐,等. 基于Fisher判别分析与随机森林的马尾松毛虫害检测[J]. 光谱学与光谱分析, 2018, 38(9): 2888-2896.
	Xu Zhang-hua, Huang Xu-ying, Lin Lu, et al. Dendrolimus punctatus walker damage detection based on fisher discriminant analysis and random forest[J]. Spectroscopy and Spectral Analysis, 2018, 38(9): 2888-2896.
10	陈妍,宋豫秦,王伟. 基于随机森林回归的草场植被盖度反演模型研究——以新疆阿勒泰地区布尔津县为例[J]. 生态学报, 2018, 38(7): 2384-2394.
	Chen Yan, Song Yu-qin, Wang Wei. Grassland vegetation cover inversion model based on random forest regression: a case study in Burqin county, Altay, Xinjiang uygur autonomous region[J]. Acta Ecologica Sinica, 2018, 38(7): 2384-2394.
11	詹国旗,杨国东,王凤艳,等. 基于特征空间优化的随机森林算法在GF-2影像湿地分类中的研究[J]. 地球信息科学学报, 2018, 20(10): 1520-1528.
	Zhang Guo-qi, Yang Guo-dong, Wang Feng-yan, et al. The random forest classification of wetland from GF-2 imagery based on the optimized feature space[J]. Journal of Geo-information Science, 2018, 20(10): 1520-1528.
12	Chen T, Trinder J, Niu R Q. Object-oriented landslide mapping using ZY-3 satellite imagery, random forest and mathematical morphology, for the Three-Gorges Reservoir, China[J]. Remote Sensing, 2017, 9: 333.
13	Zhu J, Pan Z W, Wang H, et al. An improved multi-temporal and multi-feature tea plantation identification method using Sentinel-2 imagery[J]. Sensors(Basel, Switzerland), 2019, 19(9): 2087.
14	马玥,姜琦刚,孟治国,等. 基于随机森林算法的农耕区土地利用分类研究[J]. 农业机械学报, 2016, 47(1): 297-303.
	Ma Yue, Jiang Qi-gang, Meng Zhi-guo, et al. Classification of land use in farming area based on random forest algorithm[J]. Transactions of the Chinese Society for Agricultural Machinery, 2016, 47(1): 297-303.
15	Zhou F Q, Zhang A N. Optimal subset selection of time-Series MODIS images and sample data transfer with random forests for supervised classification modelling[J]. Sensors, 2016, 16: 1783.
16	Whiteside T, Bartolo R. Mapping aquatic vegetation in a tropical wetland using high spatial resolution multispectral satellite imagery[J]. Remote Sensing, 2015, 7: 11664-11694.
17	杨艳魁,陈芸芝,吴波,等. 基于高分二号影像结合纹理信息的茶园提取[J]. 江苏农业科学, 2019, 47(2): 210-214.
	Yang Yan-kui, Chen Yun-zhi, Wu Bo, et al. Extraction of tea plantation image based on GF-2 image and texture information[J]. Jiangsu Agricultural Sciences, 2019, 47(2): 210-214.
18	Gao T, Zhu J J, Zheng X, et al. Mapping spatial distribution of larch plantations from multi-seasonal Landsat-8 OLI imagery and multi-scale textures using random forests[J]. Remote Sensing, 2015, 7: 1702-1720.
19	Wang B, Li J, Jin X F, et al. Mapping tea plantations from multi-seasonal Landsat-8 OLI imageries using a random forest classifier[J]. Journal of the Indian Society of Remote Sensing, 2019, 47: 1315-1329.
20	范敏,韩琪,王芬,等. 基于多层次特征表示的场景图像分类算法[J]. 吉林大学学报: 工学版, 2017, 47(6): 1909-1917.
	Fan Min, Han Qi, Wang Fen, et al. Scene image categorization algorithm based on multi-level features representation[J]. Journal of Jilin University (Engineering and Technology Edition), 2017, 47(6): 1909-1917.
21	郑国忠.地方政府在安吉白茶产业集群发展中的作用研究[D]. 杭州: 浙江农林大学经济管理学院, 2018.
	Zheng Guo-zhong. Study on the role of local government in the development of Anji white tea industry cluster[D]. Hangzhou: College of Economics and Management,Zhejiang Agriculture & Forestry University, 2018.
22	李光华,李俊清,张亮,等. 一种融合蚁群算法和随机森林的特征选择方法[J]. 计算机科学, 2019, 46(11): 212-215.
	Li Guang-hua, Li Jun-qing, Zhang Liang, et al. Feature selection method based on ant colony optimization and random forest[J]. Computer Science, 2019, 46(11): 212-215.
23	李平湘,刘致曲,杨杰,等. 利用随机森林回归进行极化SAR土壤水分反演[J]. 武汉大学学报:信息科学版, 2019, 44(3): 405-412.
	Li Ping-xiang, Liu Zhi-qu, Yang Jie, et al. Soil moisture retrieval of winter wheat fields based on random forest regression using quad-polarimetric SAR images[J]. Geomatics and Information Science of Wuhan University, 2019, 44(3): 405-412.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

特征集	子特征集	包含特征	特征数量
单季节初始	（1）光谱特征集	光谱波段（波段2‐7）	6
	（2）GLCM纹理特征集	6光谱波段×8个GLCM纹理特征+6个光谱波段	54
	（3）变差函数纹理特征集	6光谱波段×4个变差纹理特征+6个光谱波段	30
	（4）伪交叉变差函数纹理特征集	15对波段组合×4个伪交叉变差纹理特征+ 6光谱波段	66
	（5）植被指数特征集	3个植被指数特征+6个光谱特征	9
单季节优选	（1）GLCM纹理优选特征集	GLCM特征集前1/4	13
	（2）变差函数纹理优选特征集	变差函数纹理特征集前 1/4	7
	（3）伪交叉变差函数纹理优选特征集	伪交叉变差函数纹理特征集前1/4	16
多季节优选	（1）多季节光谱特征集	3个季节的光谱特征	18
	（2）多季节综合特征集	3个季节的综合特征（9×9窗口）	423
	（3）多季节优选特征集	多季节综合特征集前1/5特征	85

土地覆盖类型	训练集样本像素数/个	验证集样本像素数/个
茶园	6230	2259
水体	573	942
森林	10747	3095
建筑用地	1032	683
裸土	689	348
耕地	1281	658

类别		单季节初始特征集				单季节优选特征集
类别		PA/%	OM/%	OA/%	Kappa系数	PA/%	OM/%	OA/%	Kappa系数
冬季	光谱特征集	71.3	28.7	89.5	0.857	-	-	-	-
	GLCM纹理特征集	71.4	28.6	87.6	0.830	75.7	24.3	89.6	0.858
	变差函数纹理特征集	69.2	30.8	85.9	0.807	70.7	29.3	86.7	0.819
	伪交叉变差函数纹理特征集	65.2	34.8	87.0	0.823	73.5	26.5	89.3	0.855
	植被指数特征集	72.1	27.9	89.7	0.860	-	-	-	-
秋季	光谱特征集	67.6	32.4	86.2	0.812	-	-	-	-
	GLCM 纹理特征集	77.2	22.8	88.2	0.839	75.9	24.1	84.6	0.791
	变差函数纹理特征集	79.2	20.8	86.9	0.822	72.0	28.0	80.2	0.732
	伪交叉变差函数纹理特征集	79.5	20.5	87.8	0.834	78.4	21.6	87.4	0.830
	植被指数特征集	72.8	27.2	87.1	0.825	-	-	-	-
春季	光谱特征集	61.5	38.5	83.1	0.769	-	-	-	-
	GLCM 纹理特征集	74.1	25.9	85.8	0.805	66.6	33.4	81.5	0.745
	变差函数纹理特征集	77.4	22.6	85.1	0.795	64.8	35.2	78.2	0.700
	伪交叉变差函数纹理特征集	77.4	22.6	88.3	0.841	62.7	37.3	81.5	0.748
	植被指数特征集	63.2	36.8	82.5	0.761	-	-	-	-

类别	PA/%	OM/%	OA/%	Kappa
多季节光谱特征集	78.4	21.6	91.9	0.890
多季节综合特征集	86.1	13.9	99.3	0.868
多季节特征选择特征集	87.5	12.5	92.4	0.897

	PA/%	OM/%	OA/%
随机森林	87.5	12.5	92.4
U‐Net网络	86.1	13.9	90.1
最大似然法	76.5	23.5	75.8
支持向量机	83.6	16.4	89.3