基于时空融合的Landsat反射率数据时序重建与分类质量评价

Time-series Reconstruction and Classification Quality Evaluation of Landsat Reflectance Data Based on Spatio-temporal Fusion

多时相遥感数据的土地覆盖分类精度比单时相数据的分类精度更高,而中高分辨率传感器的重复观测频次低,严重制约了基于时间序列影像分类的精度。时空融合技术是解决时序观测数据缺失的有效手段,但该技术在基于时序数据的分类研究中的应用尚缺乏充分验证。针对此问题,以辽宁省部分地区为研究区,以Landsat和MODIS数据为研究对象,以STARFM、ESTARFM及半物理融合模型为年度Landsat时间序列数据的生成手段,以随机森林、最大似然及支持向量机方法为时序分类器,对比分析了不同融合模型与分类器的协同分类精度。实验结果表明:时空融合处理能够有效提升时序分类的精度尤其是植被类型地物,并且对分类器的选择不敏感。

Land cover classification based on multi-temporal remote sensing data is more accurate than that base on single temporal data.Owing to the low frequency of repeated observations by medium and high resolution sensors,the classification accuracy based on time series images is severely restricted.Spatio-temporal fusion technology is an effective method to address the lack of time series observation data,but its application in classification research based on time series data has not been fully verified.In order to solve this problem,in this article a part of Liaoning Province was taken as the research area,Landsat and MODIS data as research objects,STARFM,ESTARFM,and Semi-Physical fusion models as annual Landsat time series data generation methods,and Random Forest,Maximum Likelihood,and Support Vector Machine methods as time-series classifiers,the collaborative classification accuracy of different fusion models and classifiers.The experimental results show that the spatio-temporal fusion processing can effectively improve the accuracy of time-series classification,especially vegetation-type features,it is not sensitive to the choice of classifier.