Recently, many deep learning-based methods have been successfully applied to hyperspectral image (HSI) classification. Nevertheless, training a satisfactory network usually needs enough labeled samples. This is unfeasible in practical applications since the labeling of samples is time-consuming and expensive. The target domain samples that need to be classified are usually limited in HSIs. To mitigate this issue, a novel spectral-spatial domain attention network (SSDA) is proposed for HSI few-shot classification, which can transfer the learned classification knowledge from source domain contained enough labeled samples to target domain. The SSDA includes a spectral-spatial module, a domain attention module, and a multiple loss module. The spectral-spatial module can learn discriminative and domain invariance spectral-spatial features. The domain attention module can further enhance useful spectral-spatial features and avoid the interference of useless features. The multiple loss module, including few-shot loss, coral loss, and mmd loss, can solve the domain adaptation issue. Extensive experimental results demonstrate that on the Salinas, the University of Pavia (UP), the Indian Pines (IP), and the Huoshaoyun datasets, the proposed SSDA obtains higher classification accuracies than state-of-the art methods in the HSI few-shot classification.
The existing compressive sensing (CS) reconstruction algorithms require enormous computation and reconstruction quality that is not satisfying. In this paper, we propose a novel Dual-Channel Reconstruction Network (DC-Net) module to build two CS reconstruction networks: the first one recovers an image from its traditional random under-sampling measurements (RDC-Net); the second one recovers an image from its CS measurements acquired by a fully connected measurement matrix (FDC-Net). Especially, the fully connected under-sampling method makes CS measurements represent original images more effectively. For the two proposed networks, we use a fully connected layer to recover a preliminary reconstructed image, which is a linear mapping from CS measurements to the preliminary reconstructed image. The DC-Net module is used to further improve the preliminary reconstructed image quality. In the DC-Net module, a residual block channel can improve reconstruction quality and dense block channel can expedite calculation, whose fusion can improve the reconstruction performance and reduce runtime simultaneously. Extensive experiments manifest that the two proposed networks outperform state-of-the-art CS reconstruction methods in PSNR and have excellent visual reconstruction effects.