基于双年份三季相光谱混合分解的耕地变化遥感检测方法初步研究——以甘肃省民勤县为例

为切实保护耕地,基于遥感技术对不同耕地演化类型,包括不变耕地、新增耕地、撂荒耕地和退耕耕地进行识别和检测,以民勤县为例,采用2010和2015年2个年份的春、夏、初冬地物景观三季相特点的Landsat TM/OLI数据作为主要数据源。依据不同耕地演化类型的季相变异规律和内部要素特征选取具有物理意义的端元丰度值、丰度值差值建立2年份混合决策树划分不同耕地演化类型。同时对比分析在2010和2015年各自决策树土地利用/覆被分类结果后比较提取的耕地演化类型。结果表明:民勤县大部分耕地未发生变化,新增耕地在各个区域均有一定量的分布且主要分布在下游;退耕耕地大面积分布在坝区;撂荒耕地主要分布在沙漠绿洲过渡带。研究表明通过能够与地物先验知识关联的端元丰度值、特征端元丰度的差值建立决策树可以有效实现不同耕地演化类型的检测。误差矩阵精度评价结果显示,基于混合决策树耕地演化类型提取结果在总体精度上要高于分类后变化检测的结果。

Remote sensing detection method for the cultivated land changes based on the spectral mixture analysis of two-year and three-season phase: Taking Minqin County as an example

In order to effectively protect cultivated land,based on remote sensing technology,different types of cultivated land,including unchanged cultivated land,newly added cultivated land,abandoned cultivated land and returned cultivated land were identified and tested.Taking Minqin as an example,this study adopted Landsat TM/OLI data obtained in spring,summer and early winter of 2010 and 2015 to reflect different landscape characteristics of ground object as the main data source of this study.According to the seasonal phase variation rule of different cultivated land evolution types and the characteristics of internal factors,the endmember abundance and feature endmember abundance change with physical significance were selected to establish a two-year mixed decision tree to classify different evolution types of cultivated land.At the same time,the land evolution types by the ratio of change detection of land use/cover classification results in the respective decision tree in 2010 and 2015 were compared.The classification results showed that most cultivated land in Minqin County was changed,and the newly added cultivated land had a certain amount of distribution in all regions and mainly distributed downstream.Large areas of returning cultivated land were located in the region of Baqu.The abandoned land was mainly distributed in the transition zone of desert oasis.In conclusion,the decision tree could be established through the endmember abundance and feature endmember abundance change,which were connected with prior knowledge of ground objects effectively to detect the evolution types of different cultivated land.