论文部分内容阅读
传统的高光谱图像分类主要是基于像素的光谱特征,在一定程度上忽略了高光谱遥感图像中像素之间的空间相关性。为了充分利用高光谱图像中的空间信息,提出了一种基于加权多结构元素无偏差形态学的空间特征提取方法,并基于形态学的多尺度特征和结构保持性提出了基于邻域的多尺度空间特征提取方法,得到了高光谱遥感图像的空间特征。对k-NN分类算法进行改进,提出了基于变精度粗糙集和重构误差的k-NN分类算法,实现了基于空间特征的高光谱遥感图像分类。在两个不同的高光谱遥感图像的实验验证了基于空间特征和改进k-NN分类算法的性能。
Traditional hyperspectral image classification is mainly based on the spectral characteristics of pixels, to a certain extent, ignoring the spatial correlation between pixels in hyperspectral remote sensing images. In order to make full use of the spatial information in hyperspectral images, a spatial feature extraction method based on the unbiased morphology of weighted multi-structure elements is proposed. Based on the morphological multi-scale features and structural retentivity, a neighborhood-based multi-scale Spatial feature extraction method, the hyperspectral remote sensing image of the spatial characteristics. The k-NN classification algorithm is improved, and a k-NN classification algorithm based on variable precision rough set and reconstruction error is proposed, which achieves the classification of hyperspectral remote sensing images based on spatial features. Experiments on two different hyperspectral remote sensing images demonstrate the performance of the algorithm based on spatial features and improved k-NN classification.