首页 | 本学科首页   官方微博 | 高级检索  
     检索      

基于GF-1 WFV数据的玉米与大豆种植面积提取方法
引用本文:黄健熙,侯矞焯,苏伟,刘峻明,朱德海.基于GF-1 WFV数据的玉米与大豆种植面积提取方法[J].农业工程学报,2017,33(7):164-170.
作者姓名:黄健熙  侯矞焯  苏伟  刘峻明  朱德海
作者单位:中国农业大学信息与电气工程学院,北京,100083
基金项目:国家自然科学基金资助 (41671418, 41471342,41371326)
摘    要:准确掌握农作物的空间种植分布情况,对于国家宏观指导农业生产、制定农业政策有重要意义。针对黑龙江省玉米与大豆生育期接近、光谱特征相似,较难区分的问题,以多时相16 m空间分辨率高分一号(GF-1)卫星宽覆盖(wide field of view,WFV)影像为数据源,选择归一化植被指数(normalized difference vegetation index,NDVI)、增强植被指数(enhanced vegetation index,EVI)、宽动态植被指数(wide dynamic range vegetation index,WDRVI)、归一化水指数(normalized difference water index,NDWI)4个特征,结合实地调查样本点,采用随机森林分类算法,提取黑龙江省黑河市嫩江县玉米与大豆种植面积。研究表明,区分玉米与大豆的最佳时段为9月下旬至10月上旬,即大豆已收获而玉米未收获的时段,在4个待选特征中,NDVI、NDWI与WDRVI指数组合表现最佳;随机森林算法与最大似然算法、支持向量机算法相比,分类精度更高,其总体分类精度为84.82%,Kappa系数为77.42%。玉米制图精度为91.49%,用户精度为93.48%;大豆制图精度为91.14%,用户精度为82.76%。该方法为大区域农作物的分类提供重要参考和借鉴价值。

关 键 词:遥感  提取  分类  随机森林  大豆  玉米
收稿时间:2016/10/6 0:00:00
修稿时间:2017/4/11 0:00:00

Mapping corn and soybean cropped area with GF-1 WFV data
Huang Jianxi,Hou Yuzhuo,Su Wei,Liu Junming and Zhu Dehai.Mapping corn and soybean cropped area with GF-1 WFV data[J].Transactions of the Chinese Society of Agricultural Engineering,2017,33(7):164-170.
Authors:Huang Jianxi  Hou Yuzhuo  Su Wei  Liu Junming and Zhu Dehai
Institution:College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China,College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China,College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China,College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China and College of Information and Electrical Engineering, China Agricultural University, Beijing 100083, China
Abstract:Planting area and spatial distribution information of crops are vital for guiding agricultural production, taking effective management measurements, and monitoring crop growth conditions. Numerous crop classification algorithms have been developed with rapid development of different remote sensing data. However, distinguishing of corn and soybean cropping areas still remains a difficult challenge due to their similar growth calendar and spectral characteristics. In this study, we tried to identify corn and soybean cropping area using random forest (RF) classifier which has been proved to be an effective method in land cover classification based on multi-temporal GF-1 WFV (wide field of view) imagery. We selected Nenjiang County, Heilongjiang Province in China as the study area which was called the Town of Soybean. Seven GF-1 WFV time-series images (April 14th, May 20th, June 26th, July 16th, August 26th, September 4th, and September 29th), from which the key growth stages could be extracted and the effects of clouds could be avoided, were chosen to classify main crops. First, we conducted atmospheric and geometric corrections on multi-temporal GF-1 imagery. In order to improve the accuracy of distinguishing corn and soybean cropping area, the parameters of RF classifier were input, which included normalized difference vegetation index (NDVI), wide dynamic range vegetation index (WDRVI), enhanced vegetation index (EVI), and normalized difference water index (NDWI), and hundreds of field sample points were collected in the field survey. Also, it's necessary to evaluate the importance of different combination of these indices. The results showed that the combination of NDVI, WDRVI and NDWI achieved the best accuracy with the producer accuracy of 91.14% for soybean and 91.49% for corn, and with the user accuracy of 82.76% for soybean and 93.48% for corn. Then, the support vector machine (SVM) and maximum likelihood (ML) supervised classifiers were also used to map corn and soybean cropping areas; the classification results from the SVM and ML methods were compared with that from the RF approach with the Nenjiang Farm as the case study. The comparisons showed that the crop classification from the RF classifier had the higher accuracy than the others. Our results indicated that GF-1 data had particular advantages in mapping cropping area with its higher spatial and temporal resolutions, and could provide more effective remote sensing data during crop growth season. The temporal changes of main crops showed the best classifying date was September 29th when soybean has been harvested but corn hasn't, and their vegetation indices showed the maximum difference. The multi-temporal imagery contributed to the separation of different spectral feature curves of different crops in the growth stages when crops had similar temporal variation profiles, which helped to decrease the omission and commission errors of the resultant mapping. The results also showed that the extracted spectral information of water and construction land was very different from vegetation and could be easily masked. Comparing the SVM and ML classifiers with RF classifier, the results suggested that RF classifier could successfully distinguish corn and soybean, and its overall accuracy reached up to 84.82%. This study provides important reference for crop mapping in other agricultural regions.
Keywords:remote sensing  extraction  classification  random forest  soybean  corn
本文献已被 CNKI 万方数据 等数据库收录!
点击此处可从《农业工程学报》浏览原始摘要信息
点击此处可从《农业工程学报》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号