基于遥感指数的阳宗海流域影像分类提取

Image Classification and Extraction of Yangzong Lake Basin Based on Remote Sensing Index

基于哨兵2号MSI多光谱数据产品,利用NDVI和MNDWI遥感指数分别提取分类样本区掩膜,在掩膜上构建样本来比较6种监督学习方法的分类效果,并统计分类地物像元所占面积和比例。结果显示:(1)较为复杂的情况下,MNDWI指数提取水体信息的效果要优于NDWI指数。(2)从整体上看,马氏距离法和最大似然法可以很好地区分和识别裸土与建筑像元的特征;除平行六面体法外的其他5种分类法均对植被的分类效果表现较优;水体分类效果最好的是马氏距离法;建筑分类效果最好的是最大似然法;裸土分类效果最优的是马氏距离法。(3)马氏距离法的精度和Kappa系数均为最高,分别是98.93%、0.9767;生产者精度和用户精度最高的分别为最大似然法和马氏距离法;平行六面体法的总体精度、Kappa系数、生产者精度和用户精度均为最低。(4)在阳宗海流域地区,裸土、植被和建筑面积分别为80.28、67.82和6.82 km2,建筑面积占比最小,阳宗海流域人口密度较低。

Based on Sentinel-2 MSI multispectral data products,NDVI and MNDWI remote sensing indexes were used to extract the mask of classified sample area,and samples were constructed on the mask to compare the classification effects of six supervised learning methods,and the area and proportion of classified ground objects were counted.The results show that:(1)In more complex situations,the MNDWI index performs better than the NDWI index in water extraction.(2)Overall,the Markov distance method and maximum likelihood method can effectively distinguish bare soil and building features.Except for Parallelepiped method,other classification methods have better effect on vegetation classification.In water classification,Markov distance classification results are the best.In building classification,the maximum likelihood method has the best effect.In bare soil classification,the Markov distance method has the best effect.(3)The Mahalanobis distance method has the highest accuracy,with an overall accuracy of 98.93%and a Kappa coefficient of 0.9767.The maximum likelihood method has the highest producer accuracy,while the Markov distance method has the highest user accuracy.The overall precision,producer precision and user precision of the parallelepiped method are the lowest.(4)In Yangzong Lake drainage area,bare soil,vegetation and building area are 80.28 km2,67.82 km2 and 6.82 km2 respectively.The building area is the lowest,and the population density in Yangzong Lake Basin is low.