顾及水域的Sentinel-2数据地形阴影检测方法研究

Study on Terrain Shadow Detection of Sentinel-2 Data Considering Water Area

针对诸如水体等暗色地物信息影响山区地形阴影提取精度的问题,提出一种基于地物第一主成分特征和光谱特征的地形阴影提取方法。首先,分析包括地形阴影在内的4类典型地物光谱特征和第一主成分特征,建立阴影分量PCA1和水体分量NDMBWI,构建归一化阴影指数NSI;然后,通过分析NSI和NDVI之间的二维空间分布构建动态阈值;最后,分割图像信息,获取地形阴影区域。试验结果表明:①相较于其他方法,基于NSI的动态阈值法总体精度和Kappa系数均最高(约为0.893和0.759),阴影区域反射率的3个统计量(R、SD和CV系数)均较低,表明该方法能有效消除水体和其他暗色地物影响,准确提取阴影;②基于NSI构建的动态阈值法在不同时相和不同研究区上的地形阴影提取结果良好,地形阴影同水体、暗色地物和建筑区分度较高,能在一定程度上抑制云阴影的影响,算法具有良好的稳定性和适用性。

Aiming at the problem that dark feature information such as water bodies affects the accuracy of ter⁃rain shadow extraction in mountainous areas,this paper proposes a terrain shadow extraction method based on the first principal component features and spectral features of ground objects.Firstly,the spectral features and the first principal component features of four typical ground features including topographic shadows were ana⁃lyzed,and the shadow component(PCA1)and the water component(NDMBWI)were established to con⁃struct the Normalized Shadow Index(NSI).Then,the dynamic threshold was constructed by analyzing the two-dimensional spatial distribution between NSI and NDVI.Finally,the image information is segmented to obtain the terrain shadow area.The test results show that:(1)Compared with other methods,the dynamic threshold method based on NSI has the highest overall accuracy and Kappa coefficient(about 0.893 and 0.759).The three statistics(Range,Standard Deviation,and Coefficient of Variation)of the reflectance in the shadow area are the lower,indicating that the method can effectively remove the influence of water and other dark ground objects,and accurately extract the shadow;(2)The dynamic threshold method based on NSI can ex⁃tract topographic shadows in different phases and different study areas with good results.The topographic shad⁃ows are highly distinguishable from water bodies,dark features and buildings,and can suppress the influence of cloud shadows to a certain extent.The algorithm has good stability and applicability.