时序数据集构建质量对土地覆盖分类精度的影响研究

Influence of Time Series Data Quality on Land Cover Classification Accuracy

研究通过对MODIS双星数据组合、线性插值和HANTS平滑方法来提升时序数据集质量,采用随机森林的方法分类,对分类结果精度评定以分析时序数据集构建质量对分类精度的影响。结果表明:双星数据有利于提高时序数据集的时间分辨率,精确刻划覆盖变化,为后续处理提供基础;线性插值可改善像元点的质量,降低云、雨因素影响;HANTS平滑能移除异常值,平滑数据,突出曲线特征,降低分类复杂度。改进质量后的时序数据集,分类总体精度从84.32%提高至90.75%,Kappa系数从0.798 6提高至0.881 6。总之,使用时序数据进行土地覆盖分类时,应以消除异常值,真实反映地表覆盖物候特征为目的提高时序数据集的质量,从而提高分类精度。

This paper improved the quality of time series data sets through threemethods double star data combination of MODIS,linear interpolation and HANTS smoothing. In this study,we used random forest classification and analyzed the impact of the quality of time series dataset construction on classification accuracy though evaluating the accuracy of classification results. Results showed that the double-star data was beneficial to improve the temporal resolution of time series dataset,accurately depict the coverage change,and provide the basis for subsequent processing;linear interpolation could improve the quality of pixel points and reduce the influence of cloud and rain factors;HANTS smoothing could remove outliers,smooth data,highlight curve features,and reduce classification complexity. After improving the quality of the time series data set,the overall classification accuracy increased from 84.32% to 90.75%,and the Kappa coefficient increased from 79.86% to 88.16%. In a word,when using time series data for land cover classification,the quality of the time series data set should be improved to eliminate the outliers and truly reflect the surface covering phenological features,and the classification accuracy of the results should be improved.