【作者】 苗翠翠；

【导师】 彭世揆；

【作者基本信息】 南京林业大学 ， 森林经理学， 2010， 硕士

【摘要】 随着光谱仪器的飞速发展,它提供的多样性遥感数据同时加快了国内外对农作物提取方法的研究水平。由于MODIS数据具有免费、多时相性等优点逐渐成为国家粮食作物种植面积统计的重点研究对象。然而由于数据本身存在的大量坏数据严重制约了研究中的各项分析结果,特别对MODIS-NDVI时间序列的分析产生了很大影响,导致分类精度不高。本次研究采用一种先进的滤波处理方法—SG滤波平滑处理来改善数据的不足。从结果中可以得出：利用去除MODIS数据中噪声的方法,不会导致数据中细节信息的丢失,更重要的是经过TIMESAT处理后的影像中包含了所有地物物候期的关键信息。通过将其与NDVI时间序列结合分析,可以达到精确区分不同种作物的目的。通过对重建后的NDVI时间序列进行重点分析,将水体、建筑等典型地物进行剔除；并着重就水稻、小麦、林地等的关键物候信息进行对比研究。利用不同作物间的生长周期、生长幅度、生长长度以及生长中的NDVI最大值各不相同,采用先进的决策树进行分类规则训练,提取江苏省水稻的空间分布及种植面积。最后,通过多种手段,对分类后的面积进行定性和定量评价,得出98.6%的高精度。这不仅满足了国家对大范围内作物预测的要求,而且该方法更具有很强的实际应用价值。更多还原

【Abstract】 Along with the rapid development of spectral instrument, the multiple remote sensing data which it provided accelerates domestic and foreign to the crops extraction method research speed. With the free, composed multi-temporal MODIS images have gradually become the advantages of national grain equal emphasis of crop planting area of statistics. However, the existence of bad data itself have seriously restricted the data analysis results of the study, especially on MODIS-NDVI time series analysis has great influence, causing the classification accuracy is not high. This study adopts an advanced filter processing methods-SG smoothing filter to improve data. From the results can be obtained:the methods by removing the noise of composed multi-temporal MODIS images, won’t cause the loss of data in detail information, more important is TIMESAT after processed image contains all the key information phenology features. Through it with the NDVI time series union analysis, may achieve the precise discrimination different crops the goal.Based on the reconstruction of NDVI time series analysis, we can be on the typical residential buildings, wafer to eliminate, then we will emphatically on wheat、rice、forest on key phenology information comparative study. Using different crops growth cycle, the growth margin, growth and growing NDVI maximum length of each are not identical, Using different crops growth cycle, the growth margin, growth and growing NDVI maximum length of each are not identical, adopts advanced decision tree to carry on the classified rule training, then we can got the rice planting area and spatial distribution. Finally, through many kinds of methods carry on qualitative and the quantitative evaluation to the area, and obtains 98.6% high accuracy. Not only this has satisfied the country to the wide range in the crops forecast request, moreover this method has the very strong practical application value.更多还原