Feature extraction is of significance for hyperspectral image (HSI) classification. Compared with conventional hand-crafted feature extraction, deep learning can automatically learn features with discriminative information. However, two issues exist in applying deep learning to HSIs. One issue is how to jointly extract spectral features and spatial features, and the other one is how to train the deep model when training samples are scarce. In this paper, a deep convolutional neural network with two-branch architecture is proposed to extract the joint spectral-spatial features from HSIs. The two branches of the proposed network are devoted to features from the spectral domain as well as the spatial domain. The learned spectral features and spatial features are then concatenated and fed to fully connected layers to extract the joint spectral-spatial features for classification. When the training samples are limited, we investigate the transfer learning to improve the performance. Low and mid-layers of the network are pretrained and transferred from other data sources; only top layers are trained with limited training samples extracted from the target scene. Experiments on Airborne Visible/Infrared Imaging Spectrometer and Reflective Optics System Imaging Spectrometer data demonstrate that the learned deep joint spectral-spatial features are discriminative, and competitive classification results can be achieved when compared with state-of-the-art methods. The experiments also reveal that the transferred features boost the classification performance.

Recently, the graph convolutional network (GCN) has drawn increasing attention in the hyperspectral image (HSI) classification. Compared with the convolutional neural network (CNN) with fixed square kernels, GCN can explicitly utilize the correlation between adjacent land covers and conduct flexible convolution on arbitrarily irregular image regions; hence, the HSI spatial contextual structure can be better modeled. However, to reduce the computational complexity and promote the semantic structure learning of land covers, GCN usually works on superpixel-based nodes rather than pixel-based nodes; thus, the pixel-level spectral–spatial features cannot be captured. To fully leverage the advantages of the CNN and GCN, we propose a heterogeneous deep network called CNN-enhanced GCN (CEGCN), in which CNN and GCN branches perform feature learning on small-scale regular regions and large-scale irregular regions, and generate complementary spectral–spatial features at pixel and superpixel levels, respectively. To alleviate the structural incompatibility of the data representation between the Euclidean data-oriented CNN and non-Euclidean data-oriented GCN, we propose the graph encoder and decoder to propagate features between image pixels and graph nodes, thus enabling the CNN and GCN to collaborate in a single network. In contrast to other GCN-based methods that encode HSI into a graph during preprocessing, we integrate the graph encoding process into the network and learn edge weights from training data, which can promote the node feature learning and make the graph more adaptive to HSI content. Extensive experiments on three data sets demonstrate that the proposed CEGCN is both qualitatively and quantitatively competitive compared with other state-of-the-art methods.

Enhancing the spatial resolution of hyperspectral image (HSI) is of significance for applications. Fusing HSI with a high resolution (HR) multispectral image (MSI) is an important technology for HSI enhancement. Inspired by the success of deep learning in image enhancement, in this paper, we propose a HSI-MSI fusion method by designing a deep convolutional neural network (CNN) with two branches which are devoted to features of HSI and MSI. In order to exploit spectral correlation and fuse the MSI, we extract the features from the spectrum of each pixel in low resolution HSI, and its corresponding spatial neighborhood in MSI, with the two CNN branches. The extracted features are then concatenated and fed to fully connected (FC) layers, where the information of HSI and MSI could be fully fused. The output of the FC layers is the spectrum of the expected HR HSI. In the experiment, we evaluate the proposed method on Airborne Visible Infrared Imaging Spectrometer (AVIRIS), and Environmental Mapping and Analysis Program (EnMAP) data. We also apply it to real Hyperion-Sentinel data fusion. The results on the simulated and the real data demonstrate that the proposed method is competitive with other state-of-the-art fusion methods.

Assessing the quality of a reconstructed hyperspectral image (HSI) is of significance for restoration and super-resolution. Current image quality assessment methods such as peak signal-noise-ratio require the availability of pristine reference image, which is often not available in reality. In this paper, we propose a no-reference hyperspectral image quality assessment method based on quality-sensitive features extraction. Difference of statistical properties between pristine and distorted HSIs is analyzed in both spectral and spatial domains, then multiple statistics features that are sensitive to image quality are extracted. By combining all these statistics features, we learn a multivariate Gaussian (MVG) model as benchmark from the pristine hyperspectral datasets. In order to assess the quality of a reconstructed HSI, we partition it into different local blocks and fit a MVG model on each block. A modified Bhattacharyya distance between the MVG model of each reconstructed HSI block and the benchmark MVG model is computed to measure the quality. The final quality score is obtained by average pooling over all the blocks. We assess five state-of-the-art super-resolution methods on Airborne Visible Infrared Imaging Spectrometer (AVIRIS) and Hyperspec-VNIR-C (HyperspecVC) data using our proposed method. It is verified that the proposed quality score is consistent with current reference-based assessment indices, which demonstrates the effectiveness and potential of the proposed no-reference image quality assessment method.

Super-resolution (SR) is significant for hyperspectral image (HSI) applications. In single-frame HSI SR, how to reconstruct detailed image structures in high resolution (HR) HSI is challenging since there is no auxiliary image (e.g., HR multispectral image) providing structural information. Wavelet could capture image structures in different orientations, and emphasis on predicting high-frequency wavelet sub-bands is helpful for recovering the detailed structures in HSI SR. In this study, we propose a multi-scale wavelet 3D convolutional neural network (MW-3D-CNN) for HSI SR, which predicts the wavelet coefficients of HR HSI rather than directly reconstructing the HR HSI. To exploit the correlation in the spectral and spatial domains, the MW-3D-CNN is built with 3D convolutional layers. An embedding subnet and a predicting subnet constitute the MW-3D-CNN, the embedding subnet extracts deep spatial-spectral features from the low resolution (LR) HSI and represents the LR HSI as a set of feature cubes. The feature cubes are then fed to the predicting subnet. There are multiple output branches in the predicting subnet, each of which corresponds to one wavelet sub-band and predicts the wavelet coefficients of HR HSI. The HR HSI can be obtained by applying inverse wavelet transform to the predicted wavelet coefficients. In the training stage, we propose to train the MW-3D-CNN with L1 norm loss, which is more suitable than the conventional L2 norm loss for penalizing the errors in different wavelet sub-bands. Experiments on both simulated and real spaceborne HSI demonstrate that the proposed algorithm is competitive with other state-of-the-art HSI SR methods.

Limited by the shape-fixed kernels, convolutional neural networks (CNNs) are usually difficult to model difform land covers in hyperspectral images (HSIs), leading to inadequate land use. Recently, benefiting from the ability to conduct shape-adaptive convolutions and model complex patterns in graph-structured data, graph convolutional networks (GCNs) have been applied to HSI classification. However, due to the massive computation in GCNs, HSI is usually pretreated into a graph based on a specific superpixel segmentation, which limits the modeling of spatial topologies to the same scale. To break this limitation, we propose a multilevel superpixel structured graph U-Net (MSSGU) to learn multiscale features on multilevel graphs. Specifically, we construct several hierarchical segmentations from fine to coarse by progressively merging adjacent superpixels and then convert them into multilevel graphs. Meanwhile, based on the merging relations between hierarchical superpixels, we establish the pooling and unpooling functions to transfer features from one graph to another, thereby enabling different-level graphs to collaborate in a single network. Different from concatenating different-scale features straightforwardly in the feature fusion stage, MSSGU fuses them in a coarse-to-fine progressive manner, which can generate subtler fusion features adaptive to the pixelwise classification task. Moreover, we use a CNN instead of GCN to extract and fuse the pixel-level features, which greatly reduces the computation. Such a hybrid U-Net can exploit features of HSIs from a multiscale hierarchical perspective, and its performance has been proven competitive with other deep-learning-based methods by extensive experiments on three benchmark datasets.