论文部分内容阅读
文章为缓解农业遥感农作物分类识别方面的分类精度与成本、空间分辨率与识别方法之间的矛盾问题提供新的方法。以中分辨率的Landsat TM/ETM+遥感影像为主要数据源,农田系统图斑的矢量数据作为参考,面向对象方法和CART决策树算法等相结合,提出一种面向对象决策树的农作物分类识别方法。文章分别用提出的方法和传统的决策树分类方法对新疆焉耆盆地2011至2014年的各种地物进行分类识别,4年的平均分类精度92.09%(Kappa系数为0.91)。新方法不仅提高农作物分类识别的总精度,还能比较完整地保持农田作物的轮廓。该成果为中等分辨率的遥感影像以高精度的农作物分类识别、信息提取和动态监测提供可能。
The article provides a new method for alleviating the contradiction between classification precision and cost, agricultural spatial resolution and identification method in agricultural remote sensing crop classification and identification. Taking Landsat TM / ETM + remote sensing images with medium resolution as the main data source and the vector data of farmland system patches as a reference, the object-oriented method and the CART decision tree algorithm are combined to propose a crop identification and classification method based on object-oriented decision tree . In this paper, the proposed method and the traditional decision tree classification method are respectively used to classify the various landforms of the Yanqi Basin in Xinjiang from 2011 to 2014. The average classification accuracy in four years is 92.09% (Kappa coefficient is 0.91). The new method not only improves the overall accuracy of crop classification and identification, but also maintains the contour of cropland more completely. This result provides a possibility for high-resolution crop classification and identification, information extraction and dynamic monitoring of medium-resolution remote sensing images.