面向地理国情监测的CRF遥感影像分类

发布时间：2018-04-04 14:42

本文选题：地理国情监测　切入点：地表覆盖信息提取　出处：《应用科学学报》2016年04期

【摘要】：针对地理国情监测中地表覆盖信息的提取,提出了一种基于条件随机场的高分辨率遥感影像自动分类方法.与面向对象的传统分类方法不同,该方法基于概率图模型分别计算像素级和对象级的势函数,以及像素与它所属对象之间的层间势函数,将所得势函数统一到一个CRF模型中进行图割求解.该方法较充分地表达了像素与对象之间的关系,从而降低了对象分割误差传递对影像分类结果的影响.以"高分1号"遥感影像为实验数据,借鉴地理国情普查中地表覆盖分类体系进行实验验证.分类总体精度和平均精度分别达到91.08%和86.95%,远高于基于面向对象的分类结果.
[Abstract]:An automatic classification method of high resolution remote sensing images based on conditional random field is proposed for the extraction of ground cover information in the monitoring of geographical conditions.Different from the traditional object-oriented classification method, based on the probabilistic graph model, the potential functions at pixel level and object level are calculated, respectively, and the interlayer potential functions between pixels and objects to which they belong are calculated, respectively.The obtained potential function is unified into a CRF model for graph cutting solution.The relationship between pixels and objects is fully expressed, and the effect of object segmentation error transmission on image classification results is reduced.Taking the remote sensing image of "Gao Fen 1" as experimental data, the classification system of ground cover in the survey of geographical conditions is used for reference.The total and average accuracy of classification are 91.08% and 86.95% respectively, which is much higher than that of object-oriented classification.
【作者单位】：西安科技大学测绘科学与技术学院;武汉大学遥感信息工程学院;北京吉威时代软件股份有限公司;
【基金】：国家自然科学基金(No.41101410)资助
【分类号】：TP751

【相似文献】