【作者】 朱琳；

【导师】 郭交；

【作者基本信息】 西北农林科技大学 ， 工程硕士（专业学位）， 2018， 硕士

【摘要】 快速准确地获取农作物的空间分布、种植面积和产量等信息对国家粮食政策和经济计划具有重要作用。遥感技术是农情监测的重要手段之一,具有高效、快速的特点。目前作物分类和面积提取研究大部分建立在Landsat数据、SPOT数据和高分卫星数据等光学遥感数据和RADARSAT等雷达遥感数据上。而关于Sentinel-1雷达数据和Sentinel-2光学数据的研究还有待深入探索。本文以Sentinel-1和Sentinel-2多源遥感数据为数据源,选择最佳时相影像及有云影像,利用四种常用分类方法进行作物分类和冬小麦种植面积提取研究,分析Sentinel多源遥感数据在有云和无云情况下作物分类优势以及面积提取方面的可行性。主要得出以下结论:(1)12波段多源遥感数据分类结果比6波段多源遥感数据更好。在无云层覆盖的情况下,12波段多源遥感数据分类结果中分类精度最好达到93.47%,Kappa系数为87.07%;在有少量云层覆盖的情况下,12波段多源遥感数据分类结果最好达到90.25%,Kappa系数为81.58%。(2)多源遥感数据分类结果比光学数据更好。在无云层覆盖的情况下,BP神经网络法处理多源遥感数据数据作物分类结果提升明显,整体精度提升了3个百分点,Kappa系数提高了6个百分点;在有少量云层覆盖的情况下,同样是BP神经网络对农作物分类结果提升明显,整体分类精度和Kappa系数分别提高了7和14个百分点。(3)Sentinel多源遥感数据在冬小麦种植面积提取结果较好。在基于Sentinel多源遥感数据进行冬小麦种植面积提取结果中,最小距离法识别结果最好,识别精度达90.22%,达到了实际应用需求,得到冬小麦种植面积约为11.1131平方千米,占杨凌区总面积的8.23%。更多还原

【Abstract】 It plays an important role in national food policy and economic plan to acquire spatial distribution,planting area and yield information of crops quickly and accurately.Remote sensing technology is one of the most important ways for agricultural situation surveillance.Currently,most studies on crop classification and area extraction are based on optical remote sensing data such as Landsat data,SPOT data and high-resolution satellite data,and radar remote sensing data such as RADARSAT.Study on radar and optical remote sensing data from the newly launched Sentinel systems needs to be explored.This paper taking Sentinel-1 and Sentinel-2 multi-source remote sensing data as data source,chooses the optimal temporal data and the one with some cloud.Then,four commonly adopted classification methods are utilized to crop classification and area extraction.From the experimental results,we can see the advantages with Sentinel multi-source remote sensing data in the cases of both cloud and no cloud in crop classification.Meanwhile,the application with Sentinel multi-source remote sensing data in area extraction has been clearly proved.The main conclusions are drawn as follows:(1)Classification results of 12-band multi-source remote sensing data are better than those of the 6-band.In the case of no cloud cover,the classification accuracy of 12-band multi-source remote sensing data is 93.47%,and the Kappa coefficient is 87.07%.In the case of cloud cover,the classification accuracy of 12-band multi-source remote sensing data classification result is decreased to 90.25%,and the Kappa coefficient is only 81.58%.(2)Classification results of Sentinel multi-source remote sensing data are better than those of optical data.Without cloud cover,the crop classification results of multi-source remote sensing data by the method of BP neural network are improved significantly,and overall accuracy is improved by 3% and Kappa coefficient is raised by 6%;With a small amount of cloud cover,the crop classification results of multi-source remote sensing data by the method of BP neural network are also improved significantly.Overall classification accuracy and Kappa coefficient are increased by nearly 7% and 14%,respectively.(3)Results of winter wheat planting area extraction with Sentinel multi-source remote sensing data are better.The result of winter wheat planting area extraction based on Sentinel multi-source remote sensing data shows that the minimum distance method can get the best identification result.The accuracy of identification reaches 90.22%,which meets the practical application requirements.Finally,the winter wheat planting area is about 11.1131 square kilometers,accounting for 8.23% of the total area of Yangling.更多还原