基于像元分解的亚像元级遥感数据分类精度评估方法

A new method for assessing of remotely-sensed data the accuracy of classification based on pixel unmixing

为减小混合像元现象给遥感数据分类精度评估过程带来的影响,将像元分解技术引入到分类精度评估中,提出了基于像元分解的亚像元级分类精度评估方法,并给出了相应的处理流程及精度评估所需误差矩阵的计算方法。实验结果表明,亚像元级分类精度评估方法能够体现混合像元在分类精度评估过程中的影响,能够从一定程度上细化各分类算法的优劣,降低由于空间分辨率所引起的分类精度评估结果的不确定性。

To reduce the influence of the mixed pixel on classification accuracy assessment in remotely-sensed data, the pixel unmixing technology was introduced into the assessment study, and a new method based on pixel unmixing was proposed for the classification accuracy assessment in remotely-sensed data. The flow chart of the method and the calculation of the error matrix for classification accuracy assessment were also presented. Some experimental results show that the proposed methodology can reduce the uncertainty of classification accuracy assessment caused by low spatial resolution, and can also give more detailed assessment for different classification methods on the sub-pixel level.