The precise classification of hyperspectral remote sensing images has significant applications in both civilian and military domains.Despite the success of deep learning-based methods in hyperspectral image classification,they often lack robustness and generalization when dealing with unknown objects in open-set environments.To enhance the robustness of hyperspectral image classification methods while maintaining the accuracy of known classes,this paper proposes the spectral-spatial information reconstruction framework that simultaneously performs spectral feature reconstruction,spatial feature reconstruction,and pixel-level classification in open-set environments.By reconstructing the spectral and spatial features of hyperspectral images,the framework enhances feature representation capabilities,preserving spectral-spatial information crucial for rejecting unknown classes and distinguishing known classes.Experimental results validate the effectiveness of the proposed approach in open-set hyperspectral image classification.
维普
万方数据