A New Superpixel Segmentation Method for Hyperspectral Remote Sensing Images
In order to solve the problem of low segmentation accuracy of simple linear iterative clustering algorithm in hyperspectral remote sensing image superpixel segmentation tasks,a new unsupervised hyperspectral remote sensing image superpixel segmentation method based on multi-level linear iterative clustering combined with improved label propagation algorithm(LPA)is proposed.Firstly,we expand the applicability of Simple Linear Iterative Clustering(SLIC)to perform superpixel initial segmentation on hyperspectral images through multiple channels,and then perform multi-level iterative and detailed segmentation on superpixels with large color standard deviations.A texture feature extraction method based on local binary mode for hyperspectral remote sensing images is introduced to calculate the texture features of hyperspectral images and fuse multiple spectral features to calculate the similarity between superpixels to construct a weighted graph network,Finally,the LPA community discovery method is improved for superpixel merging,and the improved label propagation algorithm is applied to superpixel merging to obtain a more stable and accurate superpixel merging effect,im-proving the accuracy of superpixel segmentation.Compared with various methods,the proposed method has more accurate superpixel seg-mentation results for hyperspectral remote sensing images,and the superpixel edges are more closely aligned with the real boundary of land objects.It can effectively improve the problem of low accuracy in superpixel segmentation of hyperspectral remote sensing images.
国家科技期刊平台
NSTL
万方数据
