In order to solve the problem that traditional remote sensing classification method is difficult to distinguish tree species of plain afforestation, high spatial images of four different periods were selected to present the distribution of forest resource with explicit clarity. The optimal segmentation parameters were obtained by combining the calculation of change rate in variance and visual interpretation and by using a tool named ESP. Feature selection was used to reduce a large number of features to simplify the process. The region was classified by the object-based classification by using mutli-temporal images and single image respectively. Pixel-based classifications were applied to compare with the accuracy of object-based classification method. The results show that the accuracy of object-based classification in mutli-temporal images is 64%, which is better than the single image results with the accuracy of 51%. SVM and MLC reaches even lower accuracy of 49% and 43% respectively. The precision of object-based KNN classifications is better than that of pixel-based classification, indicating that the object-based classification adding mutli-temporal images has superiority in identifying those afforestation tree species in ecological landscape of forest with a complex distribution.%为解决传统遥感分类方法区分平原人工造林地树种难度较大的问题,利用4个不同时相的高空间分辨率卫星影像,基于ESP计算方差变化率并结合目视解译获取影像的最佳分割尺度;通过相关系数法筛选构建的特征,进行面向对象的多时相影像和单时相影像分类,并与基于像元分类方法进行对比分析.结果表明:基于多时相影像各类别分类精度为64%,高于单时相分类精度(51%);面向对象KNN方法的分类精度优于SVM和MLC分类方法,两者精度分别为49%和43%.在树种丰富且分布复杂的平原造林林地景观中,利用多时相遥感数据,采用面向对象分类方法用于树种精细分类更具优势.
展开▼
