论文部分内容阅读
以地处半干旱地区的北京西部山区为例,利用研究区森林类型的季相特征、已有的少部分林相图、Google Earth免费影像数据等信息选择不同坡向的相同森林类型做训练样本,通过加入其他辅助数据(海拔和坡向数据),来提高Landsat TM影像的森林类型分类精度,同时对比了基于像元和面向对象方法提取森林类型的效果。结果表明:1)就半干旱山区的森林类型划分来说,TM影像的TM4、TM5、TM4-TM2及辅助数据DEM和坡向可作为TM影像森林类型划分的最佳数据源。2)单独加入海拔信息,阔叶林的提取精度提高23%,针叶林和混交林的分类精度只提高了4%~5%;单独加入坡向信息,阔叶林的提取精度只提高21%,但是针叶林和混交林的分类精度则分别提高了13%、18%,显著优于单独加入海拔信息的效果。同时加入海拔信息和坡向信息,至少可以准确区分出约70%以上的针叶林、阔叶林和混交林。3)就本研究区而言,坡向比海拔更有效地辅助提高森林分类精度。4)就混淆矩阵数据而言,面向对象的分类方法比基于像元分类结果总体精度低3%,Kappa系数低4%,但面向对象的分类结果更加符合研究区实际情况。该研究对中分辨率影像应用于半干旱山区森林类型划分具有一定的借鉴意义。
Taking the western mountainous area of Beijing, located in the semi-arid area, as an example, the authors selected the same forest types with different slopes as the training samples based on the seasonal characteristics of the forest types in the study area, the existing small number of forest phase diagrams, the free image data of Google Earth, By adding other ancillary data (elevation and aspect data) to improve the forest type classification accuracy of Landsat TM images, the effect of extracting forest types based on pixel and object-oriented methods is also compared. The results show that: 1) TM4, TM5, TM4-TM2 and DEM and aspect data of TM images can be used as the best data source for forest type classification of TM images in the semi-arid mountainous area. 2) The elevation accuracy of the broad-leaved forest increased by 23% with the elevation information separately. The classification accuracy of the coniferous forest and the mixed forest increased by only 4% -5%. By adding the slope information alone, the extraction precision of the broad-leaved forest increased only by 21% %, But the classification accuracies of coniferous forest and mixed forest increased by 13% and 18%, respectively, which were significantly better than those obtained by adding altitude alone. At the same time by adding altitude information and aspect information, at least more than about 70% of coniferous forests, broad-leaved forests and mixed forests can be accurately distinguished. 3) For the study area, the aspect ratio is more effective than the elevation to help improve the forest classification accuracy. 4) For the purpose of confusing matrix data, the object-oriented classification method is 3% lower in overall accuracy and 4% lower in Kappa coefficient than the pixel-based classification results, but the object-oriented classification results are more in line with the actual situation in the study area. This study is of some reference to the application of mid-resolution imaging in the classification of forest types in semi-arid mountainous areas.