Many deep learning models, such as convolutional neural network (CNN) and recurrent neural network (RNN), have been successfully applied to extracting deep features for hyperspectral tasks. Hyperspectral image classification allows distinguishing the characterization of land covers by utilizing their abundant information. Motivated by the attention mechanism of the human visual system, in this study, we propose a spectral-spatial attention network for hyperspectral image classification. In our method, RNN with attention can learn inner spectral correlations within a continuous spectrum, while CNN with attention is designed to focus on saliency features and spatial relevance between neighboring pixels in the spatial dimension. Experimental results demonstrate that our method can fully utilize the spectral and spatial information to obtain competitive performance.

Gaussian mixture model (GMM) has been one of the most representative models for hyperspectral unmixing while considering endmember variability. However, the GMM unmixing models only have proper smoothness and sparsity prior constraints on the abundances and thus do not take into account the possible local spatial correlation. When the pixels that lie on the boundaries of different materials or the inhomogeneous region, the abundances of the neighboring pixels do not have those prior constraints. Thus, we propose a novel GMM unmixing method based on superpixel segmentation (SS) and low-rank representation (LRR), which is called GMM-SS-LRR. we adopt the SS in the first principal component of HSI to get the homogeneous regions. Moreover, the HSI to be unmixed is partitioned into regions where the statistical property of the abundance coefficients have the underlying low-rank property. Then, to further exploit the spatial data structure, under the Bayesian framework, we use GMM to formulate the unmixing problem, and put the low-rank property into the objective function as a prior knowledge, using generalized expectation maximization to solve the objection function. Experiments on synthetic datasets and real HSIs demonstrated that the proposed GMM-SS-LRR is efficient compared with other current popular methods.

Given the chaotic background and complex content of hyperspectral images (HSIs), the complementary information of multiple features is beneficial for HSI classification. Many multifeature-based sparse representation methods have been recently proposed. However, the large-scale dictionary problem and the optimal multifeature combination strategy have remained unsolved. For the aforementioned problem, we propose a multifeature-based discriminative label consistent K-singular value decomposition (MF-LC-KSVD) algorithm for HSI classification. In the training step of the approach, a multifeature dictionary and a combined linear classifier are obtained by jointly utilizing the LC-KSVD algorithm on the extracted features. The multifeature classifier can automatically determine the relative importance of different features for classification. The “discriminative sparse-code error” constraint in the LC-KSVD facilitates the sharing of the same spare pattern of the atoms in a special feature from the same class, thereby increasing the discrimination of the linear classifier. In the testing step, a multifeature sparse matrix is obtained by introducing the above overcomplete multifeature dictionary into multifeature-based adaptive sparse representation. The class label is then predicted by applying the classifier on multifeature sparse codes. In addition, we introduce the shape-adaptive spatial region to incorporate the information of neighbor pixels, hence further improving the classification performance. Compared with other multifeature-based sparse representation methods, our proposed MF-LC-KSVD is proven to have better classification accuracy.

Polarization imaging beyond traditional imaging offers a wide range of advantages, encompassing not only the detection of geometric shapes and surfaces, but also the measurement of physical properties. The divided focal plane (DoFP), an ideal real-time imaging, comprises a 2 × 2 linear polarization filter overlaid on a focal plane array sensor. Consequently, performing polarization demosaicing (PDM) becomes crucial to restore the missing components in the pixel data. The objective of PDM extends beyond the acquisition of intensity, and its primary goal is to minimize the estimation error of polarization characteristics. However, existing demosaicing networks often neglect the issues of polarization spatial-intensity attention and are trained similarly for generalization, thereby failing to effectively distinguish between genuine details and artifacts, which renders demosaicing vulnerable to distortion. In this work, we propose a novel generative adversarial network framework that integrates attention mechanisms and progressive discrimination for polarimetric images. The GAN-based framework introduces a global attention mechanism that enhances the interaction between spatial and polarization intensity information across different regions, resulting in improved network performance. Additionally, the framework generates artifact maps to normalize the model training process. To evaluate the effectiveness of our method, we assess its performance using full-resolution polarimetric image and compare it against traditional and recent PDM methods. The experimental results demonstrate that our proposed method achieves the highest peak signal-to-noise ratio (PSNR) and exhibits excellent visual quality for the output Stokes images when compared to existing methods.