In recent years, deep learning-based methods have been extensively utilized in remote sensing image scene classification and have achieved remarkable performance. The wide geographical coverage and resolution differences of scene images result in significant within-class diversity and between-class similarity, hindering the further improvement of classification accuracy. Attention-based methods automatically estimate the importance of local regions by learning weight assignments, which effectively enhance the feature extraction capability of the network. However, methods that solely rely on the network to automatically learn weight assignments may introduce biases in the attention calculations. By analyzing the specific contribution of local features to the key components of global semantics, we propose a collaborative dictionary learning network (CDLNet). CDLNet utilizes the collaborative representation method to decompose global features into a set of key semantic vectors to guide the attention learning process of the network. Specifically, we design a semantic summarization module (SSM), which reconstructs global semantic features by optimizing a low-redundancy dictionary. Next, we propose a global semantic attention module (GSAM), which calculates the contribution of local features to the global feature key information based on their correlation with the reconstructed key semantic set. Finally, an attention transfer loss is introduced to further enhance the attention of low-level feature maps. The experimental results on three publicly available datasets demonstrate that CDLNet can effectively improve within-class diversity and between-class similarity by optimizing the attention learning of the network, thereby achieving great promotion in comparison with state-of-the-art methods. The implementation is publicly available at https://github.com/liuofficial/CDLNet.

This article focuses on hyperspectral image (HSI) super-resolution that aims to fuse a low-spatial-resolution HSI and a high-spatial-resolution multispectral image to form a high-spatial-resolution HSI (HR-HSI). Existing deep learning-based approaches are mostly supervised that rely on a large number of labeled training samples, which is unrealistic. The commonly used model-based approaches are unsupervised and flexible but rely on handcrafted priors. Inspired by the specific properties of model, we make the first attempt to design a model-inspired deep network for HSI super-resolution in an unsupervised manner. This approach consists of an implicit autoencoder network built on the target HR-HSI that treats each pixel as an individual sample. The nonnegative matrix factorization (NMF) of the target HR-HSI is integrated into the autoencoder network, where the two NMF parts, spectral and spatial matrices, are treated as decoder parameters and hidden outputs, respectively. In the encoding stage, we present a pixelwise fusion model to estimate hidden outputs directly and then reformulate and unfold the model’s algorithm to form the encoder network. With the specific architecture, the proposed network is similar to a manifold prior-based model and can be trained patch by patch rather than the entire images. Moreover, we propose an additional unsupervised network to estimate the point spread function and spectral response function. Experimental results conducted on both synthetic and real datasets demonstrate the effectiveness of the proposed approach.

Hyperspectral image super-resolution addresses the problem of fusing a low-resolution hyperspectral image (LR-HSI) and a high-resolution multispectral image (HR-MSI) to produce a high-resolution hyperspectral image (HR-HSI). In this paper, we propose a novel fusion approach for hyperspectral image super-resolution by exploiting the specific properties of matrix decomposition, which consists of four main steps. First, an endmember extraction algorithm is used to extract an initial spectral matrix from LR-HSI. Then, with the initial spectral matrix, we estimate the spatial matrix, i.e., the spatial-contextual information, from the degraded observations of HR-HSI. Third, the spatial matrix is further utilized to estimate the spectral matrix from LR-HSI by solving a least squares (LS)-based problem. Finally, the target HR-HSI is constructed by combing the estimated spectral and spatial matrixes. In particular, two models are proposed to estimate the spatial matrix. One is a simple case that involves a LS-based problem, and the other is an elaborate case that consists of two fidelity terms and a spatial regularizer, where the spatial regularizer aiming to restrain the range of solutions is achieved by exploiting the superpixel-level low-rank characteristics of HR-HSI. Experiment results conducted on both synthetic and real data sets demonstrate the effectiveness of the proposed approach as compared to other hyperspectral image super-resolution methods.

Hyperspectral target detection (HTD) is an important issue in earth observation, with applications in both military and civilian domains. However, conventional representation-based detectors are hindered by the reliance on the unknown background dictionary, the limited ability to capture nonlinear representations using the linear mixing model (LMM), and the insufficient background-target recognition based on handcrafted priors. To address these problems, this paper proposes an interpretable representation network that intuitively realizes LMM for HTD, making nonlinear feature expression and physical interpretability compatible. Specifically, a subspace representation network is designed to separate the background and target components, where the background subspace can be adaptively learned. In addition, to further enhance the nonlinear representation and more accurately learn the coefficients, a lightweight multi-scale Transformer is proposed by modeling long-distance feature dependencies between channels. Furthermore, to supplement the depiction for target-background discrimination, a constrained energy minimization (CEM) loss is tailored by minimizing the output background energy and maximizing the target response. The effectiveness of the proposed method is demonstrated on four benchmark datasets, showing its superiority over state-of-the-art methods. The code for this work is available at https://github.com/shendb2022/HTD-IRN for reproducibility purposes.

Hyperspectral image (HSI) fusion refers to the reconstruction of a high-resolution HSI by fusing a low-resolution HSI (LR-HSI) and a high-resolution multispectral image (HR-MSI) over the same scene. Recently, researchers have proposed many approaches to handle this issue. However, most of them assume that both the spatial and spectral degradation functions are known, which are often limited or unavailable in reality. This article presents a novel model-driven deep network based on matrix decomposition, which considers spectral correlations and reasonably embeds the well-known observation models. Specifically, the proposed method decomposes the desired HSI into spectral basis and coefficients. The spectral basis can be estimated from the LR-HSI via singular value decomposition. To learn the coefficients, a learning model is constructed by merging the observation models, matrix decomposition, and sparsity into a concise single formulation. For solving the proposed model, a deep framework is built by unrolling the alternating direction method of multipliers (ADMM), dubbed as ADMM-HFNet, where the involved parameters can be learned adaptively. It is worth noting that the spectral basis cannot fully represent the desired HSI. Therefore, another model is constructed here to supplement the approximation error, which can also be embedded in the deep network. After checking on three datasets, it is found that the proposed method stands out from advanced competing techniques in both quality measures and visual effects.

Deep learning-based methods based on Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in hyperspectral image (HSI) classification. Most of these approaches are only based on 2-D CNN or 3-D CNN. It is dramatic from the literature that using just 2-D CNN may result in missing channel relationship information and using just 3-D CNN may make the model very complex. Moreover, the existing network models do not pay enough attention to extracting spectral-spatial correlation information. To address these issues, we propose a deep collaborative attention network for HSI classification by combining 2-D CNN and 3-D CNN (CACNN). Specifically, we first extract spectral-spatial features by using 2-D CNN and 3-D CNN, respectively, and then use a "NonLocalBlock" to combine these two kinds of features. This block serves as a typical spatial attention mechanism and makes salient features be emphasized. Then, we propose a "Conv Block" that is similar to the lightweight dense block to extract correlation information contained in the feature maps. Finally, we consider a deep multilayer feature fusion strategy, and thereby combine the features of different hierarchical layers to extract the strong correlated spectral-spatial information among them. To test the performance of CACNN approach, several experiments are performed on four well-known hyperspectral images. The results are compared with the state-of-the-art approaches and satisfactory performance is obtained by our proposed method.