Research On Sparse Classification Methods For Hyperspectral Images Based On Different Dictionary Priors

In recent years,hyperspectral data has been widely used in modern agriculture,military,mineral exploration and other application fields due to its richer spectral information.In these application fields,hyperspectral classification task is an important basic task and a hot topic.At the same time,compared to traditional classification methods,sparse classification method shows huge advantages in image processing and has become a commonly used method in hyperspectral classification.However,the idea of how to fully and reasonably use the relevant information of hyperspectral images to improve classification efficiency is currently a hot topic of research.At the same time,hyperspectral image data has spatial neighborhood similarity,and there are also many difficulties in combining the spatial information of hyperspectral images with spectral information.To solve these problems,this article proposes two effective hyperspectral classification methods for different prior situations,One is the Hybrid Sparse Representation Based Classification Total Variation(HSRC-TV)method based on synthetic dictionaries and TV regularization for hyperspectral images,and the other is the Hybrid Sparse Representation Based Classification Analysis Dictionary(HSRC-AN)method based on analytical dictionaries for hyperspectral images.Finally,by comparing with current traditional methods,the effectiveness of the proposed method is verified,proving that the algorithm proposed in this paper has improved classification accuracy and has certain research significance.