A Semantic Feature Extraction Method For Hyperspectral Image Classification Based On Hashing Learning

摘要

Aiming at extraction the semantic feature of hyperspectral image, a semantic feature extraction method based on supervised hashing learning is proposed in the paper. Firstly, a set of hash functions are defined based on hyperspectral target subspace constraint which take into account the locality and discriminability between classes. Secondly, a semantic subspace is obtained through discriminative learning algorithm by the label information of hyperspectral image. Finally, the sparse binary hash codes are obtained by eigenvector mapping which represents the semantic features of targets. In the method, hashing learning uses the similarity binary codes to express the similarity of the original hyperspectral spatial data, and it uses both spectral features and spatial neighborhood features, which leads to strong distinguishing ability on a certain class of hyperspectral image. The experimental on real hyperspectral images classification results show that the fusion of the extracted semantic features with the original hyperspectral features can effectively improve the classification accuracy.