自然资源学报 ›› 2019, Vol. 34 ›› Issue (5): 1079-1092.doi: 10.31497/zrzyxb.20190514

• 资源生态 • 上一篇    下一篇

基于人工神经网络多源数据融合的子像元冬油菜提取——以两湖平原为例

刘文斌(), 陶建斌(), 徐猛, 陈瑞卿, 郭洋   

  1. 地理过程分析与模拟湖北省重点实验室/华中师范大学城市与环境科学学院,武汉 430079
  • 收稿日期:2018-08-01 修回日期:2019-01-24 出版日期:2019-05-28 发布日期:2019-05-28
  • 作者简介:

    作者简介:刘文斌(1994- ),男,湖北鄂州人,硕士,研究方向为遥感影像的地学应用。E-mail: liuwenbin_ccnu@163.com

  • 基金资助:
    湖北省自然科学基金项目(2017CFB434)

A study of winter rape extraction at sub-pixel fusing multi-source data based on Artificial Neural Networks:A case study of Jianghan and Dongting Lake Plain

Wen-bin LIU(), Jian-bin TAO(), Meng XU, Rui-qing CHEN, Yang GUO   

  1. The College of Urban and Environmental Sciences, Central China Normal University/Key Laboratory of Geographical Processes and Simulation of Geographical Processes, Wuhan 430079, China
  • Received:2018-08-01 Revised:2019-01-24 Online:2019-05-28 Published:2019-05-28

摘要:

油菜是我国第五大农作物和重要的油料作物。获取油菜的种植分布信息对食用油市场的发展和粮食安全具有重要意义。两湖平原泛指包括湖北江汉平原和湖南洞庭湖平原在内的广大平原区域,是我国重要的粮棉油生产基地,“湖广熟,天下足”指的就是这一地区。由于耕地破碎,种植结构复杂,两湖平原轮作和间作的现象非常普遍,传统的遥感监测方法难以准确地获取冬油菜的空间分布。本文提出了一种基于人工神经网络ANN的子像元冬油菜提取方法,将时间序列MODIS-EVI和GF-1数据结合以提取两湖平原的冬油菜丰度信息。首先采用顺序前向选择SFS算法从时间序列MODIS-EVI数据集中进行物候特征优选;然后构建融合多源数据的ANN模型估算两湖平原的冬油菜丰度。结果表明:基于ANN方法获取的冬油菜分布具有较高的精度(ANN估算结果与GF-1和统计数据的验证精度分别为91.54%和74.70%),在利用中分辨率影像进行大尺度冬油菜精细制图方面显示出巨大潜力,可为我国冬油菜的空间分布制图和时空格局分析提供技术方法。

关键词: 人工神经网络, 冬油菜, 子像元, 时间序列MODIS, 两湖平原

Abstract:

Rape is the fifth largest crop type and an important oil crop in China. Obtaining the distribution information of rape is of great significance to the development of edible oil market and food security. Jianghan and Dongting Lake Plain are important production bases of grain, cotton and edible oil in China. The crop rotation and intercropping are very common due to the fragmented cropland fields and diversity of landscape. Traditional remote sensing monitoring methods are difficult to assess the spatial distribution and interannual variation of planting areas. In this paper, an Artificial Neural Network-based method for extracting winter rape from sub-pixels was proposed, and the MODIS and GF-1 high resolution data were combined to obtain winter rape abundance on the JPDLP. First, the Sequential Forward Selection algorithm was used to select phenological feature from the MODIS time series dataset. Then an ANN model fusing multi-source data was built to estimate winter rape abundance. The results showed that the distribution information of winter rape obtained by this method has high accuracy (the extraction accuracies of ANN modeled versus GF-1 and census data were 91.54% and 74.70%, respectively). This method showed that there is a great potential in the large-scale winter rape mapping using coarse resolution images. The results of this paper can provide technical methods for the spatial pattern evolution and spatial-temporal dynamics analysis of winter rape in China.

Key words: Artificial Neural Network, winter rape, sub-pixel, MODIS time series, Jianghan and Dongting Lake Plain