Assessing the quality of training data in the supervised classification of remotely sensed imagery: a correlation analysis
Assessing the quality of training data in the supervised classification of remotely sensed imagery: a correlation analysis
复制标题
DOI：
10.1080/14498596.2012.733616
复制
发表时间：
2012-12
影响因子：
1.9
通讯作者：
Y. Ge;Hexiang Bai;Jinfeng Wang;Feng Cao
中科院分区：
文献类型：
--
作者：
Y. Ge;Hexiang Bai;Jinfeng Wang;Feng Cao
训练数据在遥感图像的监督分类过程中起着重要的作用。其质量是影响图像分类精度的重要因素。因此，测量训练数据的质量对于分类过程和后续操作至关重要。本文讨论了一种分类前训练数据质量评估的新方法，并分别在类别和图像层面探讨了训练数据度量与分类图像度量之间的相关性。从Landsat TM图像中收集的五组样本数据用于相关性分析。实验结果表明，该方法可以有效地衡量训练数据的质量，并能在一定程度上反映使用相应训练数据集进行监督分类得到的分类图像的质量。
Training data play an important role in the supervised classification process of remotely sensed images. Its quality is an important factor affecting the accuracy of image classification. Therefore, measuring the quality of training data is essential for classification procedures and subsequent operations. This paper discusses a new method for the quality assessment of training data before the classification procedure and investigates the correlation between measures for training data and measures for classified images at category and image level, respectively. Five groups of sample data collected from a Landsat TM image were used in correlation analyses. The results demonstrate that the proposed method is valid for measuring the quality of training data and can, to some extent, reflect the quality of classified images which are obtained through supervised classification with the corresponding training dataset.