融合时间特征的高分辨率遥感影像分类  

Classification of high spatial resolution remotely sensed images by temporal feature fusion

作　　者：李亮[1] 应国伟[1] 文学虎[1] 何鑫[1] 

机构地区：[1]四川省第三测绘工程院,成都610500

出　　处：《国土资源遥感》2016年第2期91-98,共8页Remote Sensing for Land & Resources

基　　金：测绘地理信息公益性行业科研专项"卫星遥感与地面传感网一体化的湖泊流域地理国情监测关键技术研究"(编号:201512026);四川省地理国情监测工程技术研究中心资助项目"基于时序遥感影像的土地利用变化检测方法研究"(编号:GC201506);四川省测绘地理信息局科技计划项目"基于Web的四川省地理国情监测数据成果展示方法与实现"(编号:J2014ZC16)共同资助

摘　　要：为了充分利用历史矢量数据,并考虑地物类别的时间关联,提出了一种融合时间特征的高分辨率遥感影像分类方法。将历史时期矢量数据与新时期遥感影像相结合,利用二次分割获取像斑,通过支持向量机(support vector machine,SVM)算法获取像斑类别及像斑的单时期后验概率;依据历史时期及新时期像斑类别属性的关联,获取定量表达时间特征的地物类别转移概率;加权组合像斑的单时期后验概率与转移概率,采用迭代方法获取影像最终分类结果。在Quick Bird影像上的实验表明,该方法能够有效引入时间特征及先验知识,提高影像分类的精度。In order to make full use of the vector data in the historical period and the temporal relationship of the feature classes,the authors propose a classification method based on temporal feature fusion for high spatial resolution remotely sensed imagery in the paper. Image objects are generated using subdivision based on the vector data in the historical period and the remotely sensed imagery in the present period. SVM algorithm is adopted to get the initial class and posterior probability with a single period of the object. Class transition probabilities for description of temporal feature are calculated according to the class of the image objects in the historical and present periods. The iterative method is employed to get the final classification results after weighted combination of the posterior probability with a single period and the transition probability of the image objects. The experiment on Quick Bird imagery shows the proposed method can exploit the temporal feature effectively and improve the accuracy of the image classification.

关 键 词：矢量数据 遥感影像分类 像斑 支持向量机 时间特征 转移概率 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]