综合光谱纹理和时序信息的油茶遥感提取研究

doi:10.3964/j.issn.1000-0593(2023)05-1589-09

摘要
参考文献
相关文章 (15)

全文: PDF (15725 KB)
输出: BibTeX | EndNote (RIS)

摘要：具有极高营养价值且被誉为东方“橄榄油”的油茶树是我国南方地区重要经济林，我国是世界上油茶树分布最广的国家。提取油茶种植分布和面积对林业部门开展油茶的宏观管理和生产指导具有重要意义。以地处亚热带地物复杂且多山地丘陵的湖南省常宁市为研究区，该区域分布有大量农田和森林，且部分植被季节变化较大，对油茶的遥感提取带来了很大挑战。提出了基于春夏秋三期的GF-2号高分辨率卫星影像，综合植被指数、纹理特征、PCA主成分3种特征，以及春夏、春秋、夏秋、春夏秋四种不同时序组合和随机森林（RF）算法共构建了17种分类场景（S1—S17），运用随机森林（RF）、支持向量机（SVM）、最大似然（MLC）三种不同分类算法开展油茶遥感提取实验，筛选出最优特征组合、最佳分类季节与最优时序组合、最优分类方法。结果表明：仅基于光谱信息分类精度低，纹理特征的加入可大幅提升精度，而PCA对于精度的提升效果微弱；通过比较不同季节单时期的分类结果发现油茶提取精度最高的季节为夏季，夏季单时期影像在最优特征组合(S8)中油茶生产者精度（PA）为94.06%，油茶用户精度（UA）为92.57%；在分类场景S10—S17中实验发现，采用时序信息要比单时期影像有明显的精度提升，时序组合分类精度由高到低依次为：春夏秋、春夏、春秋、夏秋；综合光谱、纹理、时序信息通过随机森林（RF）、支持向量机（SVM）、最大似然（MLC）进行油茶提取，随机森林算法分类精度总体表现最好。采用春夏秋多时相遥感植被指数、纹理、PCA的随机森林方法(S17)是分类精度最高的方案，总体精度（OA）和Kappa系数分别为96.85%和0.961 0，油茶生产者精度（PA）为98.31%，油茶用户精度（UA）为94.33%；采用春夏时相遥感植被指数、纹理的随机森林方法（S10）为兼顾计算效率与精度的最优方案，总体精度（OA）和Kappa系数分别为95.62%和0.9458，油茶生产者精度（PA）为96.93%，油茶用户精度（UA）为95.09%。所提出的最佳油茶遥感提取方案能够为亚热带地区油茶及其他经济林的遥感监测提供参考。

关键词：油茶；遥感；时序；植被指数；纹理特征

Abstract：Camellia oleifera, which has high nutritional value and is known as oriental “olive oil”, is an important economic forest in southern China, and China has the widest distribution of Camellia oleifera in the world. Extracting the distribution and planting area of Camellia oleifera is significant for forestry departments to carry out macro-management and production guidance of Camellia oleifera. Changning City, Hunan Province, located in a subtropical zone with complex object features and many mountains and hills, is the study area. Many farmland and forests are distributed in this subtropic area, and some vegetation varies greatly in different seasons, which brings great challenges to remote sensing extraction of Camellia oleifera. This paper uses GF-2 high-resolution satellite images in spring, summer and autumn. Combining vegetation index, texture features, PCA principal components, and four different time series combinations in spring and summer, spring and autumn, summer and autumn, and random forest algorithm, 17 classification scenes (S1—S17) were constructed. Three classification algorithms, random forest, support vector machine and maximum likelihood, were used to carry out remote sensing extraction experiments of Camellia oleiferato select the optimal feature combination, classification season, time series combination and optimal classification method. The results show that the classification accuracy based only on spectral information is low, and the addition of texture features can greatly improve the accuracy, while PCA has a weak effect on improving the accuracy; By comparing the classification results of single-period remote sense data in different seasons, it is found that the season with the highest extraction accuracy of Camellia oleifera is summer. With the summer image of the optimal feature combination (S8), the producer accuracy of Camellia oleifera is 94.06%, and the user accuracy of Camellia oleifera is 92.57%. In the classification scenes S10—S17, it is found that the accuracy of using time series information is improved compared with that of single-period images, and the classification accuracy of time series combination from high to low is: spring, summer and autumn, spring and summer, spring and autumn, summer and autumn. Random Forest, Support Vector Machine and Maximum Likelihood are used to extract Camellia oleifera by integrating spectral, texture and time series information, and the classification accuracy of random forest algorithm is the best in general. The therandom forest method (S17) using multi-temporal remote sensing vegetation index, texture and PCA in spring, summer and autumn is the scheme with the highest classification accuracy. The overall accuracy and Kappa coefficient are 96.85% and 0.961 0 respectively, and the producer accuracy of Camellia oleifera is 98.31%, and the user accuracy of Camellia oleifera is 94.33%. The random forest method (S10) using remote sensing vegetation index and texture in spring and summeris the best scheme with calculation efficiency and classification accuracy. The overall accuracy and Kappa coefficient are 95.62% and 0.945 8, respectively. The producer and user accuracy of Camellia oleifera are 96.93% and 95.09%, respectively. The best remote sensing extraction scheme of Camellia oleifera proposed in this paper can provide a reference for remote sensing monitoring of Camellia oleifera and other economic forest extraction in subtropical areas.

Key words：Camellia oleifera; Remote sensing; Time sequence; Vegetation index; Texture features

收稿日期: 2022-03-07 修订日期: 2022-07-05

中图分类号:

TP753

基金资助: 国家重点研发计划项目（2020YFD1100200）资助

通讯作者: 李存军 E-mail: licj@nercita.org.cn

作者简介: 孟浩然，1997年生，北京市农林科学院信息技术研究中心和辽宁科技大学土木工程学院硕士研究生
e-mail: mhaorann@163.com

引用本文:

孟浩然，李存军，郑翔宇，宫雨生，刘玉，潘瑜春. 综合光谱纹理和时序信息的油茶遥感提取研究[J]. 光谱学与光谱分析, 2023, 43(05): 1589-1597.
MENG Hao-ran, LI Cun-jun, ZHENG Xiang-yu, GONG Yu-sheng, LIU Yu, PAN Yu-chun. Research on Extraction of Camellia Oleifera by Integrating Spectral, Texture and Time Sequence Remote Sensing Information. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(05): 1589-1597.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2023)05-1589-09 或 https://www.gpxygpfx.com/CN/Y2023/V43/I05/1589

[1] WANG Bin, HE Bing-hui, LIN Na, et al（王斌, 何丙辉, 林娜, 等）. Journal of Jilin University(Engineering and Technology Edition)[吉林大学学报(工学版)]，2021，52(7)：1719.
[2] CHEN Jian, WANG Wen-jun, SHENG Shi-jie, et al（陈健, 王文君, 盛世杰, 等）. Remote Sensing for Land and Resources(国土资源遥感)，2015, 27(2): 69.
[3] XU Han-ze-yu, LIU Chong, WANG Jun-bang，et al（徐晗泽宇, 刘冲, 王军邦, 等）. Journal of Geo-information Science（地球信息科学学报）, 2018, 20(3): 396.
[4] YAN En-ping, JI Yu, YIN Xian-ming, et al（严恩萍, 棘玉, 尹显明, 等）. Transactions of the Chinese Society of Agricultural Engineering（农业工程学报）, 2021, 37(16): 39.
[5] ZHANG Yue-qi, LI Rong-ping, MU Xi-han, et al(张悦琦, 李荣平, 穆西晗，等). Transactions of the Chinese Society of Agricultural Engineering（农业工程学报）, 2021, 37(17): 189.
[6] ZHAO Qing-zhan, JIANG Ping, WANG Xue-wen, et al（赵庆展, 江萍, 王学文, 等）. Transactions of the Chinese Society for Agricultural Machinery(农业机械学报), 2021, 52(11): 190.
[7] SONG Rong-jie, NING Ji-feng, LIU Xiu-ying, et al（宋荣杰, 宁纪锋, 刘秀英, 等）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报）, 2017, 48(3): 188.
[8] YANG Hong-yan, DU Jian-min, RUAN Pei-ying, et al（杨红艳, 杜健民, 阮培英, 等）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报）, 2021, 52(6): 186.
[9] ZHANG Chao, TONG Liang, LIU Zhe, et al（张超, 童亮, 刘哲, 等）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报）, 2019, 50(2): 163.
[10] LI Chang-chun, CHEN Wei-nan, WANG Yu, et al（李长春, 陈伟男, 王宇, 等）. Transactions of the Chinese Society for Agricultural Machinery（农业机械学报）2021, 52(12): 207.
[11] Yan L, Roy D P. Remote Sensing of Environment, 2014, 144(144): 42.
[12] Gomez C, White J C, Wulder M A. ISPRS Journal of Photogrammetry & Remote Sensing, 2016, 116: 55.
[13] Maus V, Cmara G, Cartaxo R, et al. IEEE Journal of Selected Topics in Applied Earth Observations & Remote Sensing, 2016, 9(8): 3729.
[14] TAO Huan, LI Cun-jun, ZHOU Jing-ping, et al(陶欢, 李存军, 周静平, 等). Journal of Natural Resources（自然资源学报）, 2018, 33(6): 1068.
[15] BAI Jia, SUN Rui, ZHANG He-lin, et al(柏佳, 孙睿, 张赫林, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2021, 37(14): 179.
[16] WANG Rui-hui, ZHONG Fei-xia, LIAO Wen-ting, et al（王瑞辉, 钟飞霞, 廖文婷，等）. Forest Science（林业科学）, 2014, 50(12): 40.
[17] Maxwell A E, Warner T A, Strager M P, et al. International Journal of Remote Sensing, 2015, 36(4): 954.