扩展的多特征遥感影像水体识别

Extended multi-feature remote sensing image water body recognition

针对传统遥感影像中水体识别精度不高且泛化能力较弱的问题,提出扩展的多特征融合分类方法。将NDWI、MNDWI等多个水体指数的光谱特征,通过最佳波段组合获取基于灰度共生矩阵的纹理特征,使用主成分分析变换后的特征图提取扩展的形态学属性剖面(EMAPs)空间特征,归一化融合组成光谱-纹理-形态学特征集,使用特征扩充算法进行重构,得到扩展的多特征数据集,利用深度置信网络(DBN)识别水体。实验结果表明,扩展的多特征数据集可有效提高水体识别模型的精度和泛化能力。

In view of the low precision and weak generalization ability of water body recognition in traditional remote sensing images,an extended multi-feature fusion classification method was proposed.The extracted features including the spectral feature with multiple water indexes such as NDWI,MNDWI,the texture feature based on gray symbiosis matrix were obtained through optimal band combination,and the spatial feature of extended profiles with morphological attribute filters(EMAPs)was extracted using the characteristic pattern of PCA transform.These features were normalized and merged into a spectral-textured-morphological feature set,the feature extension algorithm was used to reconstruct the multi-feature data set,and deep belief network(DBN)was used to realize water body recognition.Experimental results show that the extended multi-feature data set can effectively improve the precision and generalization ability of water body recognition model.