APP下载

基于时序特征的玉麦轮作制耕地提取方法研究

2021-10-12李百红彭勃董超

安徽农业科学 2021年19期

李百红 彭勃 董超

摘要 [目的]该研究对小麦、玉米轮作制下耕地的面积与分布有重要意义。[方法]基于目视判别收集样本点和GEE平台,分析地物的NDVI、EVI等指数时序特征,开发特征识别算法提取2018年玉麦轮作区面积分布,同时与CART算法监督分类结果进行精度分析。[结果]与基于多光谱的监督分类相比,基于时序特征的识别算法准确率、精准率、F1 Score和Kappa系数分别提高了0.121、0.110、0.136和0.246。[结论]基于时序特征的特征识别算法可有效识别玉麦轮作区,该算法可为轮作区农业生产提供基础数据支持。

关键词 GEE;时序;归一化指数;提取算法

中圖分类号 S 126  文献标识码 A  文章编号 0517-6611(2021)19-0214-04

doi:10.3969/j.issn.0517-6611.2021.19.056

开放科学(资源服务)标识码(OSID):

Study on Extraction Method of Wheat-Maize Rotation System Based on Time Series Characteristics

LI Bai-hong1, PENG Bo2, DONG Chao2

(1.Taian Natural Resources and Planning Bureau of Shandong Province, Taian, Shandong 270000;2.College of Information Science and Engineering, Shandong Agricultural University, Taian, Shandong 271018)

Abstract [Objective] It is important to study the area and distribution of cultivated land under wheat and maize rotation system. [Method] Based on the collection of sample points and GEE platform, we analyzed the NDVI, EVI and other index time series features of ground objects, developed feature recognition algorithm to extract the 2018 wheat-maize rotation area distribution, and conducted precision analysis with the supervised classification results of CART algorithm. [Result] Compared with the multi-spectral supervised classification, the accuracy, F1 Score and Kappa coefficient of the time-series feature recognition algorithm were improved by 0.121, 0.110, 0.136 and 0.246, respectively. [Conclusion] The feature recognition algorithm based on time series features could effectively identify wheat-maize rotation area, and the algorithm could provide basic data support for agricultural production in rotation area.

Key words GEE;Time series;Normalized index;Extraction algorithm

基金项目 山东省重点研发项目(14032761,140380198)。

作者简介 李百红(1985—),女,山东青州人,工程师,硕士,从事土地资源利用与信息化工作。*通信作者,讲师,博士,从事农业信息技术研究。

收稿日期 2021-02-26

快速准确获取作物种植的分布信息有利于实现对耕地的精确管理、产量估算、制定科学的粮食管理政策[1]。当前,作物识别多基于遥感影像光谱差异,如熊勤学等[2-3]利用NDVI提取小麦的种植面积,路中等[4-6]提取玉米的种植面积,刘佳等[7-8]完成了中国冬小麦制图和大规模春小麦提取研究。受限于传统数据处理工具的低效,这些研究只使用了一种指数提取地物信息,难以发挥高分辨率多光谱遥感数据的优势。多时相的遥感数据进行分类不仅可以提取地物物候特征,而且可有效避免单一时相存在的异物同谱和同谱异物状况[9]。……

登录APP查看全文