顾及阴影检测的BP神经网络高光谱影像分类

Hyperspectral Image Classification Based on BP Neural Network Considering Shadow Detection

针对高光谱影像中地物阴影对高光谱影像分类的影响,以桂林市为研究区域,利用珠海一号高光谱影像数据,提出一种顾及阴影检测的BP神经网络分类方法。结合研究区的归一化差值植被指数(Normalized Difference Vegetation Index,NDVI)值、归一化水指数(Normalized Difference Water Index,NDWI)值与DEM高程数据,通过决策树提取阴影区域。构建回归模型,利用函数拟合将阴影区域拟合至明亮区域,通过BP神经网络进行分类。结果表明,基于顾及阴影检测的BP神经网络分类能有效区分水体与阴影,总体精度达98.08%,Kappa系数0.968,较直接分类结果提高了0.78%,Kappa系数提高了0.013。所提方法能有效提高存在大量阴影的高光谱影像的分类效果。

Aiming at the influence of ground object shadow in hyperspectral image on hyperspectral image classification,taking Guilin as the research area,a BP neural network classification method considering shadow detection is proposed based on Zhuhai-1 hyperspectral image data.Firstly,combining the NDVI value,NDWI value and DEM elevation data of the study area,the shadow area is extracted through the decision tree.Then,the regression model is constructed,the shadow area is fitted to the bright area by function fitting,and the classification is carried out by BP neural network.The results show that the BP neural network classification based on shadow detection can effectively distinguish between water and shadow,the overall accuracy is 98.08%,kappa coefficient is 0.968,which is 0.78%higher than the direct classification results,Kappa coefficient is 0.013 higher than the direct classification results.The proposed method can effectively improve the classification effect of hyperspectral images with a large number of shadows.