地球信息科学学报 ›› 2021, Vol. 23 ›› Issue (12): 2275-2291.doi: 10.12082/dqxxkx.2021.210233

• 遥感科学与应用技术 • 上一篇    下一篇

基于高分六号宽幅影像的油菜种植分布区域提取方法

姜楠1(), 张雪红1,2,*(), 汶建龙3, 葛州徽1   

  1. 1.南京信息工程大学遥感与测绘工程学院,南京 210044
    2.河北省气象与生态环境重点实验室, 石家庄 050021
    3.61363部队,西安 710054
  • 收稿日期:2021-04-27 修回日期:2021-06-18 出版日期:2021-12-25 发布日期:2022-02-25
  • 通讯作者: *张雪红(1980— ),男,江西余干人,博士,副教授,主要从事农业遥感、气象生态遥感等研究。 E-mail: zxhbnu@nuist.edu.cn
  • 作者简介:姜 楠(1998— ),男,江西南昌人,硕士生,主要研究方向为农业遥感。E-mail: hqyjiang@163.com
  • 基金资助:
    国家自然科学基金项目(41871239);江苏高校“青蓝工程”项目;河北省省级科技计划资助(21567624H)

Extraction Method of Rapeseed Planting Distribution Area based on GF-6 WFV Image

JIANG Nan1(), ZHANG Xuehong1,2,*(), WEN Jianlong3, GE Zhouhui1   

  1. 1. School of Remote Sensing & Geomatics Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China
    2. Hebei Provincial Key Lab for Meteorology and Eco-environment, Shijiazhuang 050021, China
    3. Unit 61363, Xi'an 710054, China
  • Received:2021-04-27 Revised:2021-06-18 Online:2021-12-25 Published:2022-02-25
  • Supported by:
    National Natural Science Foundation of China(41871239);Qing Lan Project of Jiangsu Province;S&T Program of Hebei(21567624H)

摘要:

油菜作为我国主要的农业经济作物及食用油的主要来源,及时、准确地获取其种植分布信息,是全面掌握油菜种植状况、加强生产管理、优化作物种植空间格局的重要依据。高分六号(GF-6)的宽视场(Wide Field View,WFV)传感器在可见光-近红外波段基础上增设了2个红边波段、1个黄波段和1个紫波段,为油菜遥感识别提供了更加丰富的光谱信息,进而相较于蓝、绿、红、近红外4个“传统波段”的识别精度有所提升。本文以油菜开花期内两景不同时相GF-6 WFV影像拼接图像作为数据源,选择油菜生产优势区的河南省固始县为研究区,针对油菜同其他地物的“异物同谱”现象以及不同生长阶段油菜的“同物异谱”现象,利用油菜开花期独特的反射光谱特征,结合均值间标准化近距离提出了NDSI28、S34、NDSI23和NDSI46共4个光谱指数,并由此构建油菜种植区域提取的决策树模型。研究结果表明,基于4个指数组合构建的决策树模型对油菜种植分布信息的提取达到了较好的效果,总体精度为96.17%,与随机森林、支持向量机、最大似然法相比分别高出0.31%、0.88%和1.24%;制图精度方面,决策树法为98.15%,比随机森林、支持向量机、最大似然法分别高4.72%、4.21%和5.59%;对于用户精度,决策树法为86.89%,较随机森林、最大似然法分别低2.2%和1.63%,比支持向量机高0.11%。由此说明,GF-6 WFV数据中的新增波段极大地丰富了其光谱信息,使其在包括油菜在内的农作物种植分布信息提取中具有独特的优势和巨大潜力。

关键词: 高分六号, 宽幅影像, 油菜, 决策树分类, 光谱指数, 多生育期, 种植分布, 固始县

Abstract:

Rapeseed is the main agricultural cash crop and the main source of cooking oil in China. Timely and accurate acquisition of the spatial distribution of rape plants is important for understanding the status of rape planting, strengthening production management, and optimizing the spatial pattern of crop planting. The Wide Field View (WFV) sensor of Gaofen-6 (GF-6) adds a purple band, a yellow band, and two red edge bands to the visible-near-infrared bands, which provides more spectral information for rape identification from remote sensing, thus improving the identification accuracy compared with the "traditional bands" of blue, green, red and near-infrared. In this paper, the Gushi county, Henan province, a dominant area of rape, was selected as the research area. Two GF-6 WFV images within the flowering period of rape were mosaiced as the data source. Due to the phenomenon of "same spectrum with different species" between rape and other land objects and the phenomenon of "same species with different spectrum" for rape at different growth phases, we put forward four spectral indices including NDSI28、S34、NDSI23, and NDSI46, based on the unique spectral reflectance characteristics of rape at flowering phase and the algorithm of standardized close range between means. A decision tree model for rape identification was then constructed based on these indices. The results show that the decision tree model based on the combination of four indices achieved a high accuracy in extracting rape, with an overall accuracy of 96.17%, which was 0.31%, 0.88%, and 1.24% higher than that of random forest, Support Vector Machine(SVM), and maximum likelihood method, respectively. The cartography accuracy of decision tree model was 98.15%, which was 4.72%, 4.21%, and 5.59% higher than that of random forest, SVM, and maximum likelihood method, respectively. The user accuracy of the decision tree model was 86.89%, which was 2.2% lower than that of random forest, 1.63% lower than that of the maximum likelihood method, and 0.11% higher than that of the SVM. The cartographic accuracy of different classification methods was greater than 90%, and in particular, the decision tree model showed the highest cartographic accuracy. In terms of the user accuracy, the random forest showed the highest value (89.09%), the Support Vector Machine (SVM) showed the lowest value (86.78%), and the decision tree method showed an user accuracy of 86.89%. As a result, the new bands in the GF-6 WFV data can greatly enrich the spectral information of rape. Our results demonstrated the unique advantages and great potential of the new bands in the GF-6 WFV data in the extraction of crop planting region and distribution information, including rape.

Key words: GF-6, WFV, rapeseed, decision tree classification, spectral index, multiple growth periods, planting distribution, Gushi County