基于优选特征及月合成Landsat数据湿地提取研究

Study on Wetland Extraction Using Optimal Features and Monthly Synthetic Landsat Data

针对Landsat卫星完整时间序列数据难以获取导致的湿地提取准确性较低和湿地提取最优特征不明确的问题,该文提出一种基于优选特征和月合成时间序列Landsat数据提取湿地的方法。通过月合成方法,利用Landsat7ETM+和Landsat8OLI数据构建Landsat 30m地表反射率、NDVI、NDWI和缨帽变换湿度分量的时间序列;利用随机森林算法和扩展的Jeffries-Matusita距离(JBh)优选对湿地提取贡献较大的特征,并基于优选特征提取湿地。结果显示:1)月合成方法有效地改善了条带和云覆盖造成的Landsat单景影像数据缺失问题;2)5月NDVI和6、8月NDWI以及5月TC-Wetness是区分永久性草本沼泽、水稻田、草地和旱地的最优特征;3)基于优选特征的湿地分类结果总体精度达到0.91,Kappa系数为0.89。特征优选减少了数据冗余,提高了运算效率,为提高湿地分类精度提供了理论基础。

Wetland mapping is an important tool for understanding wetland functions and monitoring their response to natural and anthropogenic actions,but mapping wetland distribution at 30 m resolution is difficult because：the Landsat time series data are generally irregular due to the missing value caused by cloud cover,and optimal features for wetland identification are still not clear.In this paper,a wetland extraction workflow was proposed.Firstly,Landsat7 ETM＋ and Landsat8 OLI data were merged to generate monthly Landsat 30 m NDVI, NDWI and TC-Wetness time series.The contribution of each feature for wetland classification was evaluated using extension method of Jeffries-Matusita distance （J Bh ） and the random forest （RF） importance score,and the optimal features for wetland identification were then selected.Finally,the wetland was identified using the optimal features and the RF classifier.The results show that：1） Monthly synthetic Landsat 30 m time series data can make up the missing data caused by cloud cover;2） NDVI in May,NDWI in June and August and TC-Wetness in May can distinguish permanent marsh,paddy fields,grassland and dry farmland effectively;3） The overall accuracy and Kappa coefficient using optimal features were 0.91 and 0.89,respectively.It is found that the wetland identification accuracy can be improved effectively based on the method proposed in this paper.