Search

Recent studies have shown that graph‐based clustering methods are good at processing hyperspectral images (HSIs), while falling short for large‐scale HSIs due to high time complexity. Meanwhile, the performance of these methods relies on the quality of the constructed graph with selected anchor points. More anchor points bring better clustering results for graph‐based methods. Time complexity, however, sees a considerable increase as the number of anchor points grows. Therefore, a method that can obtain efficient clustering accuracy and consumes less time is to be developed. Against this backdrop, a novel algorithm named fast spectral clustering via efficient multilayer anchor graph (FEMAG) is proposed to resolve the accuracy and time‐consuming trade‐off problem. First, FEMAG adopts superpixel principal component analysis (SuperPCA) to extract the low‐dimensional features of HSIs. Then, a multilayer anchor graph is constructed to improve the clustering performance. When constructing the similarity graph, FEMAG takes balanced K‐means‐based hierarchical K‐means (BKHK) to obtain outperforming anchor points efficiently. Extensive experiments validate that FEMAG achieves better clustering accuracy while taking less time compared to previous clustering methods. [ABSTRACT FROM AUTHOR]

Hyperspectral images (HSIs) possess an inherent three-order structure, prompting increased interest in extracting 3D features. Tensor analysis and low-rank representations, notably truncated higher-order SVD (T-HOSVD), have gained prominence for this purpose. However, determining the optimal order and addressing sensitivity to changes in data distribution remain challenging. To tackle these issues, this paper introduces an unsupervised Superpixelwise Multiscale Adaptive T-HOSVD (SmaT-HOSVD) method. Leveraging superpixel segmentation, the algorithm identifies homogeneous regions, facilitating the extraction of local features to enhance spatial contextual information within the image. Subsequently, T-HOSVD is adaptively applied to the obtained superpixel blocks for feature extraction and fusion across different scales. SmaT-HOSVD harnesses superpixel blocks and low-rank representations to extract 3D features, effectively capturing both spectral and spatial information of HSIs. By integrating optimal-rank estimation and multiscale fusion strategies, it acquires more comprehensive low-rank information and mitigates sensitivity to data variations. Notably, when trained on subsets comprising 2%, 1%, and 1% of the Indian Pines, University of Pavia, and Salinas datasets, respectively, SmaT-HOSVD achieves impressive overall accuracies of 93.31%, 97.21%, and 99.25%, while maintaining excellent efficiency. Future research will explore SmaT-HOSVD's applicability in deep-sea HSI classification and pursue additional avenues for advancing the field. [ABSTRACT FROM AUTHOR]

The need for enhancing image spatial resolution has motivated the researchers to propose numerous super-resolution (SR) techniques, including those developed specifically for hyperspectral data. Despite significant advancements in this field attributed to deep learning, little attention has been given to evaluating the practical value of super-resolved images in specific applications. Most methods are validated in application-independent scenarios, often using simulated low-resolution images, resulting in overly optimistic conclusions. In this article, we propose task-based evaluation strategies for hyperspectral image SR and we present results obtained with various approaches that include pansharpening, multispectral–hyperspectral data fusion, and single-image SR. We demonstrate that the proposed framework allows us to highlight both benefits and limitations of each method and can, therefore, guide the development of SR techniques suitable for real-world applications.

Hyperspectral images (HSIs) are extensively utilized in several fields due to their abundant spectral band, particularly for tasks like ground object classification and environmental monitoring. However, as a result of equipment and imaging condition constraints, HSI frequently demonstrates a restricted spatial resolution. The fusion of a low-resolution HSI and a high-resolution multispectral image (HR-MSI) of the same scene is a crucial method for generating an HR-HSI. At present, due to factors, such as complexity and GPU memory limitation, most of the HSI–MSI fusion algorithms based on deep learning (DL) cannot utilize the transformer module well to capture the long-range dependence information in large-size remote sensing images. At the same time, the lack of a large amount of high-quality training data has become an important problem that affects the performance of fusion algorithms based on DL. In response to the above issues, this article introduces a new unsupervised multiscale transformer fusion (UMTF) network, called UMTF-Net, which enables HSI–MSI fusion without the need for additional training data. UMTF-Net is composed of an HSI fusion network and a U-network (U-Net)-based multiscale feature extraction network. In order to learn the cross-feature spatial similarity and long-range dependency of MSI and HSI, we first extract the multiscale features of MSI using the U-Net-based multiscale feature extraction network. We then input these features into the corresponding scale cross-feature fusion transformer module in the HSI fusion network to conduct feature fusion. Then, we input the fused features into the spatial spectral fuse attention module for spatial spectral feature enhancement, and finally generate HR-HSI. Comparing UMTF-Net to other advanced methods, the fusion results from three datasets and multiple ablation experiments indicate that our method performs excellently in different evaluations.

Elevation information from light detection and ranging (LiDAR) data relieve the phenomenon of “same spectrum with different object” in hyperspectral images (HSI) classification. Consequently, HSI and LiDAR joint classification is a significant research topic. However, existing methods encounter several challenges. Primarily, there exists a deficiency in intraclass information interaction and underutilization of discriminative feature. Furthermore, the process of labeling samples is time-consuming and laborious. To solve the aforementioned issues, a classification method based on pseudolabeled contrastive learning is proposed to exploit substantial amounts of unlabeled information in order to enhance intraclass information interaction. The proposed method is divided into two stages for semisupervised classification. In the first stage, an unsupervised feature extraction network is designed to improve the interaction of features from multimodal data. A multimodal data cross-attention module is proposed to enhance the interaction of multimodal information at corresponding locations. Exploiting pseudolabeling contrastive learning module facilitates the interaction of information between intraclass objects. In the second stage, supervised classification with a limited number of labeled samples is performed. The multisource discriminatively consolidate feature module is designed to generate discriminative features, which are used to guide the fusion feature enhancement process. Apart from this, this module leverages multiscale features to expand the receptive field. Tested on both self-constructed and public datasets, the proposed method provides higher classification accuracy than some existing methods with a limited amount of labeled samples.

Anomaly detection (AD) aiming to locate targets distinct from the surrounding background spectra remains a challenging task in hyperspectral applications. The methods based on low-rank decomposition utilize the inherent low-rank characteristic of hyperspectral images (HSIs), which has attracted great interest and achieved many advances in recent years. In order to fully consider the characteristics of HSIs, more appropriate constrains need to be added to the low-rank model. However, there are too many regularizations and mutual constraints between regularizers, which would result in a reduction in detection accuracy, while an increasing number of tradeoff parameters complicates parameter tuning. To address the above problems, we propose a novel method based on merging total variation into low-rank representation (MTVLRR) for hyperspectral AD in this article, using a regularizer to reflect the low-rankness and smoothness of the background component of HSIs simultaneously, which can significantly decrease the mutual influence of regularizers and the difficulty of parameter tuning. Experimental results on both simulated and real hyperspectral datasets demonstrate that the proposed MTVLRR has an excellent AD performance in terms of detection accuracy compared with other state-of-the-art methods.

When oil spills occur at sea, effective detection and monitoring are required to establish a successful response plan, and remote sensing emerges as an appropriate technology to support this process. Furthermore, Earth observation is progressively relying on low-cost multispectral sensors developed for monitoring particular features across various scenarios. However, despite these advancements, the ongoing challenge lies in reducing computational costs, resource requirements, and energy consumption. This work aims to select the spectral index that best detects coastal spills among those documented. The long-term goal is to develop a low-cost multispectral sensor with suitable bands. For this purpose, this study uses data from different sensors that acquired data of the Deepwater Horizon accident in the Gulf of Mexico. The confusion matrices, accuracy, and F1-score generated from the kNN pixel classification based on the indices values are measures of its performance. For this study, the recently introduced normalized difference oil index (NDOI) proves to be the best option for identifying coastal spills since it minimizes the false positives related to suspended sand and is quick to calculate. In addition, it has demonstrated that it helps distinguish different spill thicknesses and estimates the oil volume. Therefore, future research will focus on developing and validating a low-cost multispectral system that uses the NDOI bands to detect spills.

Deep learning methodologies have demonstrated considerable effectiveness in hyperspectral anomaly detection (HAD). However, the practicality of deep learning-based HAD in real-world applications is impeded by challenges arising from limited labeled data, large-scale hyperspectral images (HSIs), and constrained computational resources. In light of these challenges, this article introduces a convolutional neural network (CNN)-based HAD model through the incorporation of Tucker decomposition, named TD-CNND. Drawing inspiration from transfer learning, the proposed model initially constructs pixel sample pairs from known labeled HSIs in the source domain, feeding them into the designed CNN to train the network learning spectral feature differences to obtain a CNN containing knowledge from the source domain. Subsequently, to prevent the need for network retraining caused by structural changes and to reduce model parameters for improving detecting timeliness, a general network compression scheme based on Tucker decomposition is applied to the CNN, where the convolutional layers of the above CNN undergo Tucker tensor decomposition to compress the network and alleviate parameter redundancy. Finally, spectral features realignment is used to recover the detection accuracy loss caused by Tucker tensor decomposition. In addition, a dual-windows structure is implemented during the detection phase, incorporating spatial information to the aforementioned spectral-level learning model, facilitating spectral-spatial collaborative HAD. Experimental evaluations using three real hyperspectral datasets and artificially expanded datasets demonstrate that, in comparison with state-of-the-art methods, the proposed TD-CNND method exhibits effectiveness and superiority in terms of both time cost and detection accuracy, where the notable advantages in terms of time cost become more pronounced with an increasing number of pixels.

Change detection on hyperspectral images (HSIs) is an essential task for Earth observation. Due to the vast amounts of remote sensing (RS) Big Data resulting from the ongoing advancements in RS hardware, employing centralized learning through cloud computing emerges as a logical and convenient solution. Nevertheless, this approach overlooks the influence of the isolation and heterogeneity of RS data on the reliability of change detection outcomes. In contrast, federated learning (FL) enables collaborative change detection on nonindependent identical distributed RS data without the need to transfer the original data. Simultaneously, it is important to acknowledge that the dependency of FL on the central node may pose potential data security risks. To address this issue, this article proposes a decentralized FL generative adversarial network (GAN) network. This ensures that raw data remains stationary during participation in unsupervised learning. In addition, the network employs edge devices to implement FL, allowing adaptation to diverse devices in practical scenarios. Lastly, blockchain is integrated into a decentralized architecture for the dynamic selection of leader nodes, effectively enhancing the robustness and security of the framework. Decentralized federated generative adversarial network (DFGAN) is a novel approach for cooperative privacy preservation introduced into hyperspectral change detection systems. The method outperforms most of existing methods on three datasets, achieving more than 90% detection accuracy on all of them. In addition, by simulating in a real on-orbit environment using a satellite constellation simulator on a Jeston TX2, FedGAN demonstrates superior accuracy in HSI change compared to existing algorithms while significantly reducing training time by 17.3%.

Large-scale hyperspectral image (HSI) clustering remains a fundamental and challenging task due to tremendous spatial scales, abundant spectral band information, and lack of prior information. Most existing clustering methods either ignore the spectral band correlation leading to low clustering performance nor unprocessable due to the large spatial scale. To solve these difficulties, this article presents a dual smooth graph convolutional clustering (DSGCC) framework for large-scale HSI clustering. Specifically, the superpixel is introduced to decrease the spatial scale of HSI and reduce graph node number for subsequent network training. Furthermore, a smooth graph filter is presented, which extracts the smooth features and filters the high-frequency interference in graph learning. In addition, we propose a layerwise graph reconstruction mechanism, which constrains all hidden layers output by graph reconstruction loss. Finally, we introduce a self-training method that utilizes soft labels to supervise during clustering and learns the robust embedding for node clustering. DSGCC is an end-to-end network that is optimized by joint loss, which is easily trained from scratch. We assess DSGCC on five commonly used large-scale HSI datasets, and experiments denote that DSGCC achieves the optimum clustering performance, which is superior to existing HSI clustering approaches. On Salinas, Indian Pines, Pavia University, WHU-Hi-LongKou, and WHU-Hi-HongHu datasets, the clustering overall accuracy of DSGCC are 85.43%, 65.45%, 70.48%, 87.30%, and 77.34%, respectively.

Spectral indices are of fundamental importance in providing insights into the distinctive characteristics of oil spills, making them indispensable tools for effective action planning. The normalized difference oil index (NDOI) is a reliable metric and suitable for the detection of coastal oil spills, effectively leveraging the visible and near-infrared (VNIR) spectral bands offered by commercial sensors. The present study explores the calculation of NDOI with a primary focus on leveraging remotely sensed imagery with rich spectral data. This undertaking necessitates a robust infrastructure to handle and process large datasets, thereby demanding significant memory resources and ensuring scalability. To overcome these challenges, a novel cloud-based approach is proposed in this study to conduct the distributed implementation of the NDOI calculation. This approach offers an accessible and intuitive solution, empowering developers to harness the benefits of cloud platforms. The evaluation of the proposal is conducted by assessing its performance using the scene acquired by the airborne visible infrared imaging spectrometer (AVIRIS) sensor during the 2010 oil rig disaster in the Gulf of Mexico. The catastrophic nature of the event and the subsequent challenges underscore the importance of remote sensing (RS) in facilitating decision-making processes. In this context, cloud-based approaches have emerged as a prominent technological advancement in the RS field. The experimental results demonstrate noteworthy performance by the proposed cloud-based approach and pave the path for future research for fast decision-making applications in scalable environments.

Spectral indices are of fundamental importance in providing insights into the distinctive characteristics of oil spills, making them indispensable tools for effective action planning. The normalized difference oil index (NDOI) is a reliable metric and suitable for the detection of coastal oil spills, effectively leveraging the visible and near-infrared (VNIR) spectral bands offered by commercial sensors. The present study explores the calculation of NDOI with a primary focus on leveraging remotely sensed imagery with rich spectral data. This undertaking necessitates a robust infrastructure to handle and process large datasets, thereby demanding significant memory resources and ensuring scalability. To overcome these challenges, a novel cloud-based approach is proposed in this study to conduct the distributed implementation of the NDOI calculation. This approach offers an accessible and intuitive solution, empowering developers to harness the benefits of cloud platforms. The evaluation of the proposal is conducted by assessing its performance using the scene acquired by the airborne visible infrared imaging spectrometer (AVIRIS) sensor during the 2010 oil rig disaster in the Gulf of Mexico. The catastrophic nature of the event and the subsequent challenges underscore the importance of remote sensing (RS) in facilitating decision-making processes. In this context, cloud-based approaches have emerged as a prominent technological advancement in the RS field. The experimental results demonstrate noteworthy performance by the proposed cloud-based approach and pave the path for future research for fast decision-making applications in scalable environments. [ABSTRACT FROM AUTHOR]

For hyperspectral image change detection (HSI-CD) task, unsupervised learning methods based on pseudo-labels obtained from pre-classification method are promising due to its simplicity and generality. However, the inevitable noise and the lack of diversity in the pseudo-labels would degrade the performance of the network model, resulting in insufficient mining of fine change information in HSIs. To address this issue, we propose a progressive pseudo-label (PPL) framework for HSI-CD. Differing from the state-of-the-art methods of training designed networks with fixed pseudo-labels, we focus our attention on improving the quality of pseudo-labels progressively. In the PPL framework, a multi-scale pre-classification module is constructed for generating initial pseudo-labels. Subsequently, an uncertainty-aware selection strategy is designed to update the pseudo-labels. Specifically, the new pseudo-labels are selected based on the uncertainty estimated by the Bayesian network trained with pseudo-labels. The new pseudo-labels have richer diversity with quality assurance and thus called progressive pseudo-labels. Meanwhile, the Bayesian network trained by progressive pseudo-labels has increasingly better detection capability. As a result, the PPL framework can detect strong and subtle changes while suppressing false changes. Furthermore, the PPL framework additionally provides the uncertainty to evaluate the reliability of the prediction. We have implemented intensive experiments on three publicly available datasets demonstrate the outperformance and generalization of the PPL framework. The overall accuracy (OA) and Kappa metrics on the commonly used Hermiston dataset reached 0.9558 and 0.8728, respectively.

Hyperspectral images (HSIs) possess an inherent three-order structure, prompting increased interest in extracting 3D features. Tensor analysis and low-rank representations, notably truncated higher-order SVD (T-HOSVD), have gained prominence for this purpose. However, determining the optimal order and addressing sensitivity to changes in data distribution remain challenging. To tackle these issues, this paper introduces an unsupervised Superpixelwise Multiscale Adaptive T-HOSVD (SmaT-HOSVD) method. Leveraging superpixel segmentation, the algorithm identifies homogeneous regions, facilitating the extraction of local features to enhance spatial contextual information within the image. Subsequently, T-HOSVD is adaptively applied to the obtained superpixel blocks for feature extraction and fusion across different scales. SmaT-HOSVD harnesses superpixel blocks and low-rank representations to extract 3D features, effectively capturing both spectral and spatial information of HSIs. By integrating optimal-rank estimation and multiscale fusion strategies, it acquires more comprehensive low-rank information and mitigates sensitivity to data variations. Notably, when trained on subsets comprising 2%, 1%, and 1% of the Indian Pines, University of Pavia, and Salinas datasets, respectively, SmaT-HOSVD achieves impressive overall accuracies of 93.31%, 97.21%, and 99.25%, while maintaining excellent efficiency. Future research will explore SmaT-HOSVD’s applicability in deep-sea HSI classification and pursue additional avenues for advancing the field.

Recently, autoencoder (AE)-based anomaly detection approaches for hyperspectral images (HSIs) have been extensively proposed; however, the reconstruction accuracy is susceptible to the anomalies and noises. Moreover, these AE-based anomaly detectors simply compress each pixel into a hidden-layer with a lower dimension and then reconstruct it, which does not consider the spatial properties among pixels. To solve the above issues, this paper proposes a background reconstruction framework via a 3D-transformer (3DTR) network for anomaly detection in HSIs. The experimental results on both synthetic and real hyperspectral datasets demonstrate that the proposed 3DTR network is able to effectively detect most of the anomalies by comprehensively considering the spatial correlations among pixels and the spectral similarity among spectral bands of HSIs. In addition, the proposed method exhibits fewer false alarms than both traditional and state-of-the-art (including model-based and AE-based) anomaly detectors owing to the adopted pre-detection procedure and the proposed novel patch-generation method in this paper. Moreover, two ablation experiments adequately verified the effectiveness of each component in the proposed method. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have achieved great progress in the classification of surface objects with hyperspectral data, but due to the limitations of convolutional operations, CNNs cannot effectively interact with contextual information. Transformer succeeds in solving this problem, and thus has been widely used to classify hyperspectral surface objects in recent years. However, the huge computational load of Transformer poses a challenge in hyperspectral semantic segmentation tasks. In addition, the use of single Transformer discards the local correlation, making it ineffective for remote sensing tasks with small datasets. Therefore, we propose a new Transformer layered architecture that combines Transformer with CNN, adopts a feature dimensionality reduction module and a Transformer-style CNN module to extract shallow features and construct texture constraints, and employs the original Transformer Encoder to extract deep features. Furthermore, we also designed a simple Decoder to process shallow spatial detail information and deep semantic features separately. Experimental results based on three publicly available hyperspectral datasets show that our proposed method has significant advantages compared with other traditional CNN, Transformer-type models. [ABSTRACT FROM AUTHOR]

Hyperspectral image (HSI) processing tasks frequently rely on spatial–spectral total variation (SSTV) to quantify the local smoothness of image structures. However, conventional SSTV only considers a sparse structure of gradient maps computed along the spatial and spectral dimensions while neglecting other correlations. To address this limitation, we introduce low-rank guided SSTV (LRSTV), which characterizes the sparsity and low-rank priors of the gradient map simultaneously. First, we verify through numerical tests and theoretical analyses that the gradient tensors are not only sparse but also low-rank. Subsequently, to model the low rankness of the gradient map, we use the tensor average rank to represent the low Tucker rank of gradient tensors. The convex envelope of the tensor average rank is then employed to penalize the rank on the gradient map after the Fourier transform along the spectral dimension. By naturally encoding the sparsity and low-rank priors of the gradient map, LRSTV results in a more accurate representation of the original image. Finally, we demonstrate the effectiveness of LRSTV by integrating it into the HSI processing model, replacing conventional SSTV, and testing it on two public datasets with nine cases of mixed noise and two datasets with realistic noise. The results show that LRSTV outperforms conventional SSTV in terms of accuracy and robustness.

Hyperspectral image (HSI) and multispectral image (MSI) fusion has the potential to significantly improve the quality and usefulness of data, leading to better decision-making and a more complete understanding of the observed scene. For HSI and MSI fusion, capturing matched pairs of HSI and MSI images is challenging. This hampers the pretraining of neural-network-based HSI–MSI fusion methods and yields unsatisfactory fusion results. A lightweight-attention (LA) cyclic network (Circle-Net) without pretraining using labeled data is constructed and applied to HSI–MSI fusion to alleviate this issue. Circle-Net consists of a coordinate feature fusion (CFF) network and a dual-attention decoder (DAD) network. Multiscale features collected from the DAD network are fused by the CFF network to derive a high-resolution HSI. Specifically, in the DAD network, skip connections in the encoder–decoder network are replaced by LAs, while polarized attention is used to guarantee efficient transfer of features between the encoder and decoder. In comparison with other methods, the experimental performance shows the superiority of the Circle-Net in both visual and quantitative performance.

The low rank and sparse representation (LRSR) technique has attracted increasing attention for hyperspectral anomaly detection (HAD). Although a large quantity of research based on LRSR for HAD is proposed, the detection performance is still limited, due to the unsatisfactory dictionary construction and insufficient consideration of global and local characteristics. To tackle the above-mentioned concern, a novel HAD method, termed dual collaborative constraints regularized low-rank and sparse representation via robust dictionaries construction, is proposed in this article. Concretely, a robust dictionary construction strategy, which thoroughly excavates the potential of the density estimation model and local outlier factor, is proposed to yield pure and representative dictionary atoms. To fully exploit the global and local characteristics of hyperspectral images, dual collaborative constraints corresponding to the background and anomaly components are imposed on the LRSR model. Notably, two weighted matrices are further exerted on the representation coefficients to improve the effect of collaborative constraints, considering the fact that the surrounding pixels similar to the testing pixel should be given a large weight, otherwise the weight is expected to be small. In this way, the background and anomaly components can be well modeled. Additionally, a nonlinear transformation operation, which combines the output of the density estimation model and local outlier factor with the detection result derived from the LRSR model, is developed to suppress the background. The experiments conducted on one simulated dataset and three real datasets demonstrate the superiority of the proposed method compared with the four typical methods and four state-of-the-art methods.

Sparse representation (SR)-based approaches and collaborative representation (CR)-based methods are proved to be effective to detect the anomalies in a hyperspectral image (HSI). Nevertheless, the existing methods for achieving hyperspectral anomaly detection (HAD) generally only consider one of them, failing to comprehensively exploit them to further promote the detection performance. To address the issue, a novel HAD method, which integrates both SR and CR, is proposed in this article. To be specific, an SR model, whose overcomplete dictionary is generated by means of the density-based clustering algorithm and superpixel segmentation method, is first constructed for each pixel in an HSI. Then, for each pixel in an HSI, the used atoms in SR model are sifted to form the background dictionary corresponding to the CR model. To fully exploit both SR and CR information, we further combine the residual features obtained from both SR and CR model by the nonlinear transformation function to generate the response map. Finally, to preserve contour information of the objects, a postprocessing operation with guided filter is imposed into the response map to acquire the detection result. Experiments conducted on simulated and real datasets demonstrate that the proposed SRCR outperforms the state-of-the-art methods.

Traditional convolutional neural networks (CNNs) can be applied to obtain the spectral-spatial feature information from hyperspectral images (HSIs). However, they often introduce significant redundant spatial feature information. The octave convolution network is frequently utilized instead of traditional CNN to decrease spatial redundant information of the network and extend its receptive field. However, the 3D octave convolution-based approaches may introduce extensive parameters and complicate the network. To solve these issues, we propose a new HSI classification approach with a multi-scale spectral-spatial network-based framework that combines 2D octave and 3D CNNs. Our method, called MOCNN, first utilizes 2D octave convolution and 3D DenseNet branch networks with various convolutional kernel sizes to obtain complex spatial contextual feature information and spectral characteristics, separately. Moreover, the channel and the spectral attention mechanisms are, respectively, applied to these two branch networks to emphasize significant feature regions and certain important spectral bands that comprise discriminative information for the categorization. Furthermore, a sample balancing strategy is applied to address the sample imbalance problem. Expansive experiments are undertaken on four HSI datasets, demonstrating that our MOCNN approach outperforms several other methods for HSI classification, especially in scenarios dominated by limited and imbalanced sample data. [ABSTRACT FROM AUTHOR]

Due to the limitations of imaging systems, the swath width of available hyperspectral images (HSIs) is relatively narrow, making it difficult to meet the demands of various applications. Therefore, an appealing idea is to expand the width of HSIs by using widely covered multispectral images (MSIs), called spectral super-resolution (SSR) of MSIs. According to the radiation transmission process of the imaging system, the spectral mixing characteristics of ground objects can be described by the linear spectral mixing model (LSMM). Inspired by the linear mixed part and nonlinear residual part of the LSMM, we propose a double-branch SSR network. To generate wide HSIs, a spectral mixing branch is designed to extract abundances from wide MSIs and adaptively learn hyperspectral endmembers from narrow HSIs. Furthermore, considering the nonlinear factors in the imaging system and atmospheric transmission, a nonlinear residual branch is built to complement the spectral and spatial details. Finally, the SSR result can be obtained with the fusion of linear and nonlinear features. To make the network structure achieve corresponding physical significance, we constrain the network through joint loss functions at different stages. In addition to two simulated datasets with limited coverage, our model is also evaluated on a real MSI–HSI dataset in a larger area. Extensive experiments show the superiority of the proposed model compared with state-of-the-art baselines. Moreover, we visualized the internal results of our network and conducted ablation experiments on a single branch to further demonstrate its effectiveness. In the end, the influence of network hyperparameters, including endmembers and loss function weight coefficient, is discussed and analyzed in detail. [ABSTRACT FROM AUTHOR]

Maize is susceptible to infect pest disease, and early disease detection is key to preventing the reduction of maize yields. The raw data used for plant disease detection are commonly RGB images and hyperspectral images (HSI). RGB images can be acquired rapidly and low-costly, but the detection accuracy is not satisfactory. On the contrary, using HSIs tends to obtain higher detection accuracy, but HSIs are difficult and high-cost to obtain in field. To overcome this contradiction, we have proposed the maize spectral recovery disease detection framework which includes two parts: the maize spectral recovery network based on the advanced hyperspectral recovery convolutional neural network (HSCNN+) and the maize disease detection network based on the convolutional neural network (CNN). Taking raw RGB data as input of the framework, the output reconstructed HSIs are used as input of disease detection network to achieve disease detection task. As a result, the detection accuracy obtained by using the low-cost raw RGB data almost as same as that obtained by using HSIs directly. The HSCNN+ is found to be fit to our spectral recovery model and the reconstruction fidelity was satisfactory. Experimental results demonstrate that the reconstructed HSIs efficiently improve detection accuracy compared with raw RGB image in tested scenarios, especially in complex environment scenario, for which the detection accuracy increases by 6.14%. The proposed framework has the advantages of fast, low cost and high detection precision. Moreover, the framework offers the possibility of real-time and precise field disease detection and can be applied in agricultural robots. [ABSTRACT FROM AUTHOR]

Recently, autoencoder (AE)-based anomaly detection approaches for hyperspectral images (HSIs) have been extensively proposed; however, the reconstruction accuracy is susceptible to the anomalies and noises. Moreover, these AE-based anomaly detectors simply compress each pixel into a hidden-layer with a lower dimension and then reconstruct it, which does not consider the spatial properties among pixels. To solve the above issues, this paper proposes a background reconstruction framework via a 3D-transformer (3DTR) network for anomaly detection in HSIs. The experimental results on both synthetic and real hyperspectral datasets demonstrate that the proposed 3DTR network is able to effectively detect most of the anomalies by comprehensively considering the spatial correlations among pixels and the spectral similarity among spectral bands of HSIs. In addition, the proposed method exhibits fewer false alarms than both traditional and state-of-the-art (including model-based and AE-based) anomaly detectors owing to the adopted pre-detection procedure and the proposed novel patch-generation method in this paper. Moreover, two ablation experiments adequately verified the effectiveness of each component in the proposed method.

Hyperspectral images (HSIs) are usually corrupted by various types of mixed noises, which degrades the qualities of acquired images and limits the subsequent applications. In this article, we propose a novel denoised method based on a hybrid spatial–spectral total variation (SSTV) regularization, which we refer to as ${l_{0}}$-${l_{1 - 2}}$SSTV. Specifically, ${l_{0}}$-${l_{1 - 2}}$SSTV can be treated as an integrated regularization embedding spatial–spectral ${l_{0}}$ gradient model into ${l_{1 - 2}}$SSTV. ${l_{1 - 2}}$SSTV regularization exploits the sparse structures in both spatial and spectral domains. Hence, the correlations within HSIs are fully considered. Due to the good performance of ${l_{1 - 2}}$-norm in image restoration, ${l_{1 - 2}}$SSTV gives a tighter approximation for the gradient domains of HSIs. It can effectively avoid artifacts and oversmoothing caused by the limitation of the SSTV regularization based on ${l_{1}}$-norm (${l_{1}}$SSTV). Meanwhile, the ${l_{0}}$ gradient regularization controls the number of nonzero gradients to promote the local piecewise smoothness, making denoised images preserve clear edges. With the effective combination of ${l_{1 - 2}}$SSTV and ${l_{0}}$ gradient regularization, ${l_{0}}$-${l_{1 - 2}}$SSTV produces high-quality restoration results in the denoising process: better detail preservation and sharper edges. The augmented Lagrangian method and the difference of convex algorithm are exploited to optimize the proposed model. The results for simulated and real experiments demonstrate the effectiveness and superiority of the proposed method compared with state-of-the-art methods.

Multimodally distributed data is very common in remote sensing images, such as hyperspectral images (HSIs). It is important to capture the local manifold structure while preserving the global discriminant information in the multimodal data. In this article, an adaptive local discriminant analysis and distribution matching (ALDADM) method is designed for the domain adaptation (DA) in HSI classification. ALDADM uses adaptive local discriminant analysis to extract features that are discriminative and robust to multimodally distributed data. Meanwhile, considering the spectral properties of HSI, the domain shift is reduced by distribution matching and subspace alignment. During this learning process, ALDADM only selects density peak samples to reduce the influence of interfering samples. Four DA tasks are designing using University of Pavia, Center of Pavia, Yancheng, and Botswana HSI datasets. Compared with existing DA methods, the experimental results demonstrate that our ALDADM offers a superior performance.

Fusion technology has been the core strategy to obtain a high-spatial-spectral resolution hyperspectral image (HSI). In recent years, few fusion models focused on exploiting the underlying manifold structure in the spatial dimension of the high-resolution HSI. We assume that image patches of high-resolution multispectral image (HR-MSI) and high-resolution hyperspectral image (HR-HSI) share similar manifold structures. Through this bridge, a local-nonlocal low-dimensional manifold defined from HR-MSI is built to favor the patch relationship in HR-HSI, which is further utilized to build a set of orthogonal bases. Subsequently, we introduce a truncation operation based on the adaptively constructed orthogonal bases, which can efficiently preserve the low-frequency information of the image and reduce the interference of noise. Finally, we combine the local-nonlocal manifold term and the truncation operation, coined as LNTM, into a variational super-resolution framework to regularize the latent HR-HSI, leading to a strongly convex function regarding the fusion problem. The cost function is solved by a carefully designed variant of alternating direction method of multipliers algorithm. Different experiments on three public benchmarks validate our algorithm outperforms the recent start-of-the-arts concerning both visual quality and numerical metrics.

Hyperspectral images (HSIs) have always played an important role in remote sensing applications. Anomaly detection has become a hot spot in HSI processing in recent years. The popular detecting method is to accurately segment anomalies from the background. Informative bands are very important for the accuracy improvement of the detection technology. However, most of the abnormal targets segmentation methods focus on the usage of all the spectral features, thus are easily affected by redundant bands or feature noise. A hyperspectral anomaly detection algorithm based on subfeature ensemble is proposed in this article. The proposed method consists of the following steps. First, the bands of the original HSI are normalized and randomly divided into several subfeature sets according to different proportions. Second, six methods including the prior-based tensor approximation algorithm (PTA), Reed–Xiaoli method, a low-rank and sparse representation method, a low-rank and sparse matrix decomposition-based Mahalanobis distance method, the graph and total variation regularized low-rank representation-based method, and a method based on tensor principal component analysis are applied to detect anomalies on the original HSI, and the method with the best performance is used to obtain an enhanced feature set. Then, the enhanced features and the subfeatures are ensembled iteratively to construct a new dataset. Finally, the PTA method is operated on the dataset with ensemble features to get the final abnormal target results. Six hyperspectral datasets are used in the experiment. Seven methods are employed as comparisons. The results are analyzed from both qualitative and quantitative perspectives. Extensive experimental results illustrate that the proposed method performs best on all datasets.

Hyperspectral anomaly detection is a widely used technique for exploring target of interest in hyperspectral images (HSIs). In recent years, the low-rank and sparse-decomposition-based anomaly detection model has attracted extensive attention. However, these models suffer from two main problems. First, it is difficult for them to completely separate the low-rank background and the sparse anomaly. Moreover, the extracted sparse component is inevitably contaminated by noise. Second, the incorporation of various constraints increases the cost of selecting the optimal parameters. To solve the two key problems, we propose a self-adaptive low-rank and sparse decomposition (SLaSD) method for hyperspectral anomaly detection in this article. The proposed method decomposes the sparse (anomaly) part of the HSI through a novel self-adaptive alternating direction method (S-ADM). The noise of the sparse part is suppressed through a dual strategy of integrating guided filter and the difference between the S-ADM-derived sparse features of pixels. The performance of the proposed method is evaluated by comparing with ten state-of-the-art methods using six real HSIs. It is shown that the proposed SLaSD method can produce more accurate detection results than the ten benchmark methods.

Hyperspectral imaging technology has been popularly applied in remote sensing because it collects echoed signals from across the electromagnetic (EM) spectrum and thereby contributes fruitfully spatial-spectral information. However, the processing or storage of high-data-volume hyperspectral images (HSIs), also viewed as snapshots varying with the EM spectrum, burdens the hardware resources, especially for the high spectral resolution and spatial resolution cases. To address this challenge, a novel unsupervised dimensionality reduction method based on the dynamic mode decomposition (DMD) algorithm is proposed to analyze hyperspectral data. This method decomposes the spatial-spectral HSIs in terms of spatial dynamic modes and corresponding spectral patterns. Then, these spatial-spectral patterns are combined to reconstruct the raw HSIs via a low-rank model. Furthermore, we extend the proposed DMD method to hyperspectral data in the tensor form and title it CubeDMD to actualize the compression of HSIs in horizontal, vertical, and spectral dimensions. Our proposed data-driven scheme is benchmarked by the real hyperspectral data measured at the Salinas scenes and Pavia University. It is demonstrated that the HSIs can be reconstructed accurately and effectively by the proposed low-rank model. The mean peak signal-to-noise ratio between the reconstructed and original HSIs can reach 31.47 dB, and the corresponding mean spectral angle mapper is only 0.1037. Our work provides a useful tool for the analysis of HSIs with a low-rank representation.

A significant challenge in methods for anomaly detection (AD) in hyperspectral images (HSIs) is determining how to construct an efficient representation for anomalies and background information. Considering the high-order structures of HSIs and the estimation of anomalies and background information in AD, this article proposes a kernel minimum noise fraction transformation-based background separation model (KMNF-BSM) to separate the anomalies and background information. First, spectral-domain KMNF transformation is performed on the original hyperspectral data to fully mine the high-order correlation between spectral bands. Then, a BSM that combines the outlier removal, the iteration strategy, and the Reed–Xiaoli detector (RXD) is proposed to obtain accurate anomalous and background pixel sets based on the extracted features. Finally, the anomalous and background pixel sets are used as input for anomaly detectors to improve the background suppression and anomaly detection capabilities. Experiments on several HSIs with different spatial and spectral resolutions over different scenes are performed. The results demonstrate that the KMNF-BSM-based algorithms have better target detectability and background suppressibility than other state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Convolutional neural networks (CNNs) have achieved great progress in the classification of surface objects with hyperspectral data, but due to the limitations of convolutional operations, CNNs cannot effectively interact with contextual information. Transformer succeeds in solving this problem, and thus has been widely used to classify hyperspectral surface objects in recent years. However, the huge computational load of Transformer poses a challenge in hyperspectral semantic segmentation tasks. In addition, the use of single Transformer discards the local correlation, making it ineffective for remote sensing tasks with small datasets. Therefore, we propose a new Transformer layered architecture that combines Transformer with CNN, adopts a feature dimensionality reduction module and a Transformer-style CNN module to extract shallow features and construct texture constraints, and employs the original Transformer Encoder to extract deep features. Furthermore, we also designed a simple Decoder to process shallow spatial detail information and deep semantic features separately. Experimental results based on three publicly available hyperspectral datasets show that our proposed method has significant advantages compared with other traditional CNN, Transformer-type models.

The very high spectral resolution in hyperspectral images (HSIs) offers an opportunity to detect subtle land-cover changes. However, the availability of HSIs acquired from different platforms requires the development of change detection (CD) methods capable of processing HSIs with different spatial resolutions. In this article, we propose a general end-to-end subpixel convolution-based residual network (SPCNet) to accomplish the CD task between high spatial resolution (HR) and low spatial resolution (LR) HSIs. To effectively tackle the resolution matching issue, a super-resolution (SR) block with an efficient subpixel convolution layer is introduced to upscale the LR feature maps into HR maps. The subpixel convolution layer can fully explore the subpixel context information by learning an array of upscaling filters. Moreover, the designed SPC module is embedded into the LR branch to generate more discriminative representations. More importantly, the SPC module as a plug-and-play unit has the potential to be embedded into other baseline networks to enhance the feature learning capability. Experimental results on four HSI datasets demonstrate the effectiveness of the proposed SPCNet. [ABSTRACT FROM AUTHOR]

Sparse unmixing (SU) algorithms use the existing spectral library as prior knowledge to analyze the endmembers and estimate abundance maps. The majority of SU algorithms use loss functions based on the $L_{2,1}$ -norm or $F$ -norm to minimize reconstruction error. They have different advantages and shortcomings. In short, $F$ -norm has a differentiable characteristic, and it is easy to minimize as a loss function. However, it is very sensitive to heavy noise and outliers. While the $L_{2,1}$ -norm emphasizes the reconstruction error on each band and is robust to noise with different intensities in different bands. But the $L_{2,1}$ -norm is nondifferentiable at zero-point. This article introduces an adaptive loss function based on the $\sigma $ -norm for SU, which combines the advantages of the $L_{2,1}$ -norm and $F$ -norm. The adaptive loss function is related to a nonnegative parameter $\sigma $. By adjusting the parameter $\sigma $ , the adaptive loss function can approach the $F$ -norm or $L_{2,1}$ -norm. To the best of our knowledge, it is the first time to apply an adaptive loss function to SU. Moreover, the adaptive loss function is globally differentiable, and we propose an optimization algorithm for the adaptive loss function and verify its convergence. Experiments on real-world and synthetic HSIs show that the adaptive loss function effectively enhances the performance of the SU algorithms. [ABSTRACT FROM AUTHOR]

Deep learning-based methods (e.g., convolutional neural network (CNN)-based methods) have shown increasing potential in hyperspectral image (HSI) change detection (CD). However, the recent advances in CNN-based methods in HSI CD tasks are mostly devoted to designing more complex architectures or adding additional hand-designed blocks. This increases the number of parameters making model training difficult. In this article, we propose an end-to-end residual self-calibrated network (RSCNet) to increase the accuracy of HSI CD. To fully exploit the spatial information, the proposed RSCNet method adaptively builds interspatial and interspectral dependencies around each spatial location with fewer extra parameters and reduced complexity. Moreover, the introduced self-calibrated convolution (SCConv) helps to generate more discriminative representations by heterogeneously exploiting convolutional filters nested in the convolutional layer. The designed RSC module can explicitly incorporate richer information by introducing response calibration operation. The experiments on four bitemporal HSI datasets demonstrated that the proposed RSCNet method is more accurate than ten widely used benchmark methods. [ABSTRACT FROM AUTHOR]

Band selection can reduce information redundancy and improve application efficiency of hyperspectral images (HSIs), such as classification. Traditional band selection methods usually weight different objectives and combine them together in a single function, making it difficult to balance conflict between various criteria. Recently, multiobjective optimization band selection (MOBS) techniques have got a lot of attention, which evaluate bands from different aspects separately and search for a solution well-balancing all the objectives simultaneously. However, few algorithms are focused on target detection, and most of them use just two objectives, which are not sufficient enough to select optimal band subset for detection. To alleviate these problems, this article proposes an algorithm named target-oriented multiobjective optimization of band selection (TOMOBS). First, the multiobjective optimization (MO) framework with three objective functions is modeled, involving information, noise, and correlation of the bands, respectively. Second, to optimize the proposed model, the noninferior solution advantage matrix (NAM) is designed based on the swarm intelligence optimization method, which can provide accurate solutions and improve the descriptiveness of MO problems (MOPs). Third, a target-oriented evaluation mechanism is developed to guide selecting final result from the Pareto front (PF), especially designed for target detection. Experiments on real hyperspectral datasets show that this algorithm can provide a subset of bands with strong representational capability for target detection and achieve impressing results compared with the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Change detection (CD) with hyperspectral images (HSIs) can be effectively performed using deep learning networks (DLNs) by taking advantage of HSIs for their abundant spectral and spatial information and the excellent performance of DLN in machine learning. By modeling the temporal dependence of multiscale representative features, more discriminative information reflecting land use and land cover (LULC) changes can be obtained by suppressing less correlated information while improving the robustness of pseudo-changes caused by imaging noises. However, preserving time-dependent multiscale representative features while extracting spatial–spectral features based on conventional DLN is difficult, mainly due to the structural limitation of conventional DLN. A multipath convolutional long short-term memory (LSTM) multipath convolutional long short-term memory neural network (MP-ConvLSTM) taking advantage of LSTM and convolutional neural network (CNN) through the designed parallel architecture to learn multilevel temporal dependencies of bitemporal HSIs, therefore, was proposed for extracting multiscale temporal–spatial–spectral features by combining hidden states from different paths of ConvLSTM in the present study. In the proposed MP-ConvLSTM, the efficient channel attention (ECA) module was introduced to refine features of different paths, and Siamese CNN was adopted to reduce HSIs’ dimensionality and extract preliminary features to build up an end-to-end trainable model for CD with HSIs. The validity of the MP-ConvLSTM was evaluated using the binary and multiclass CD datasets. The CD accuracy of the proposed MP-ConvLSTM was visually and statistically evaluated by different criteria and compared with those derived from several state-of-the-art (SOTA) CD algorithms. The experiments demonstrated that our proposed model not only outperformed those SOTA CD models but also exhibited better tradeoff between complexity and accuracy in general. [ABSTRACT FROM AUTHOR]

Traditional convolutional neural networks (CNNs) can be applied to obtain the spectral-spatial feature information from hyperspectral images (HSIs). However, they often introduce significant redundant spatial feature information. The octave convolution network is frequently utilized instead of traditional CNN to decrease spatial redundant information of the network and extend its receptive field. However, the 3D octave convolution-based approaches may introduce extensive parameters and complicate the network. To solve these issues, we propose a new HSI classification approach with a multi-scale spectral-spatial network-based framework that combines 2D octave and 3D CNNs. Our method, called MOCNN, first utilizes 2D octave convolution and 3D DenseNet branch networks with various convolutional kernel sizes to obtain complex spatial contextual feature information and spectral characteristics, separately. Moreover, the channel and the spectral attention mechanisms are, respectively, applied to these two branch networks to emphasize significant feature regions and certain important spectral bands that comprise discriminative information for the categorization. Furthermore, a sample balancing strategy is applied to address the sample imbalance problem. Expansive experiments are undertaken on four HSI datasets, demonstrating that our MOCNN approach outperforms several other methods for HSI classification, especially in scenarios dominated by limited and imbalanced sample data.

Spectral unmixing expresses the mixed pixels existing in hyperspectral images as the product of endmembers and their corresponding fractional abundances, which has been widely used in hyperspectral imagery analysis. However, the endmember spectra even for pixels from the same material of an image may include variability due to the influence of lighting conditions and inherent properties of materials within different pixels. Though the in situ spectral library has been used to accommodate such variability by using multiple in situ spectra to represent each kind of material, the performance improvement may be restricted due to the limited number of endmembers for each material. Therefore, in this article, spectral variability is directly extracted from an in situ endmember library and considered to be transferable among different endmembers for the first time. Furthermore, such a spectral variability is further used to augment sparse unmixing by synchronously performing endmember-based reconstruction and spectral variability-augmented reconstruction in the sparse unmixing model. By, respectively, imposing sparse and smoothness regularization over abundances and variability coefficients, a convex optimization-based spectral variability augmented sparse unmixing (SVASU) is finally proposed, and its convergence performance is also analyzed. Experiments conducted over synthetic and real-world datasets demonstrate that the proposed SVASU method not only significantly improves the unmixing performance of conventional spectral library-based unmixing but also outperforms several state-of-the-art sparse unmixing algorithms. [ABSTRACT FROM AUTHOR]

Recently, deep-learning-based feature extraction (FE) methods have shown great potential in hyperspectral image (HSI) processing. Unfortunately, it also brings a challenge that the training of the deep learning networks always requires large amounts of labeled samples, which is hardly available for HSI data. To address this issue, in this article, a novel unsupervised deep-learning-based FE method is proposed, which is trained in an end-to-end style. The proposed framework consists of an encoder subnetwork and a decoder subnetwork. The structure of the two subnetworks is symmetric for obtaining better downsampling and upsampling representation. Considering both spectral and spatial information, 3-D all convolution nets and deconvolution nets are used to structure the encoder subnetwork and decoder subnetwork, respectively. However, 3-D convolution and deconvolution kernels bring more parameters, which can deteriorate the quality of the obtained features. To alleviate this problem, a novel cost function with a sparse regular term is designed to obtain more robust feature representation. Experimental results on publicly available datasets indicate that the proposed method can obtain robust and effective features for subsequent classification tasks. [ABSTRACT FROM AUTHOR]

Hyperspectral images (HSIs) contain rich spectral signatures that reveal more image details and, thus, enable the detection of less noticeable changes on the ground. However, HSI-based change detection (CD) is susceptible to a large amount of irrelevant or noisy spectral and spatial information due to massive spectral bands. To address these issues, we propose a novel spectral and spatial attention network (S2AN) for HSI-based CD, which is capable to suppress CD-irrelevant spectral and spatial information via adaptive spectral and spatial attention mechanisms. S2AN takes as input the image patch from the difference map between two HSIs and outputs the status of change for the patch. Specifically, S2AN is composed of several repeated attention blocks, each of which contains the spectral attention (SpeA) module for directly calculating the attention score for each input channel, the Gaussian spatial attention (GSpaA) module that first constructs an adaptive Gaussian distribution and then samples it to derive the attention scores for each spatial position, and the convolutional feature extraction (CFE) module for extracting features from the attention-weighted input. It is worth mentioning that, in addition to the advantage of the attention, GSpaA also reduces the sensitivity of patch size for patch-based methods. To effectively train S2AN when facing insufficient labeled data, a semisupervised strategy that combines supervised and unsupervised methods to augment labeled training data is proposed. Experiments on several HSI datasets in comparison to existing methods show the superiority of S2AN. [ABSTRACT FROM AUTHOR]

Remote sensing images, especially hyperspectral images (HSIs), are extremely vulnerable to random noise and stripe noise. As a key aspect of HSI data quality improvement, stripe noise removal has always been a pervasive issue in remote sensing image processing. Convolutional neural networks have been applied for HSI data destriping. However, the existing methods lose the stripe-free component of the original image to a certain extent. These models also ignore the global spatial context of images and the correlation between spatial information and spectral information. Therefore, we propose a novel destriping convolutional network to overcome the problems with the existing methods. Octave convolution is used to extract cross-frequency features, and separate and compress the low-frequency information of the images, while dilation convolution (Dila-Conv) is used to reduce the amount of required calculation and also preserve the key image information. In addition, Dila-Conv can expand the receptive field to obtain multiscale features. Finally, a cross-channel enhanced spatial–spectral feature fusion module is used to acquire and integrate spatial context information and interchannel dependencies on a global scale as auxiliary information so that the network model can learn and pay attention to key feature information, specifically, “what to look for” and “where to look at,” which can facilitate the distinction between stripe and stripe-free components. Experimental results obtained using multiple datasets demonstrated that the proposed method can outperform the existing comparable methods and can produce satisfactory results in terms of visual effects and quantitative evaluation. [ABSTRACT FROM AUTHOR]

Hyperspectral image (HSI) denoising is a crucial preprocessing task to improve the performance of the subsequent HSI interpretation and applications. With recent progress in deep learning, HSI denoising methods based on deep neural networks have attracted increasing interest and achieved the state-of-the-art performance. Nevertheless, most of these methods are based on network structures originally developed for grayscale and color images and require a change of network structure to be applicable to HSIs. The new network architectures often lead to complicated models and limited flexibility, which, in turn, result in difficulty in learning and demand of a large number of training samples. In this article, we propose an innovative two-stage learning method including pretraining and fine-tuning procedures. In the first stage, a denoising convolutional neural network can be pretrained with pairs of corrupted and clean images. In the second stage, the pretrained network is fine-tuned via a self-supervised learning strategy to capture the spectral correlation in HSIs. The training pairs in the second stage are constructed from the neighboring band images in the target noisy HSI, leading to a novel idea of embedding spectral information into denoiser through the target image, rather than the change of the network architecture. This model has strong adaptability such that many image denoising networks can be easily adopted for HSIs, while the external hyperspectral training set is optional but not mandatory. Experimental results show that our method has competitive performance compared with the state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Hyperspectral image (HSI) change detection aims to identify the differences in multitemporal HSIs. Recently, a graph convolutional network (GCN) has attracted increasing attention in the field of remote sensing due to its advantages in processing irregular data. In comparison with a convolutional neural network (CNN) that can only perform convolution operations on data with the assumption of the Euclidean structure, GCN adopts a graph structure to flexibly capture the characteristics and structure information of non-Euclidean data. In this article, we propose a novel dual-branch difference amplification GCN (D2AGCN) for HSI change detection with limited samples, which allows the network to fully extract and effectively amplify the difference features of multitemporal HSIs for change detection. The dual-branch structure can effectively extract sufficient different features to facilitate the detection of the changed areas. As far as we know, this is the first time that GCN has been introduced into HSI change detection. A difference magnification module is designed to suppress similar regions and highlight the feature differences between the multitemporal HSIs in the dual-branch structure, which increases the distinction between change and nonchange classes. The visual and quantitative experimental results on three real hyperspectral datasets (i.e., China, Bay Area, and Santa Barbara) show that the proposed D2AGCN outperforms most of the state-of-the-art methods in HSI change detection with limited training samples. [ABSTRACT FROM AUTHOR]

Due to the presence of multiple scatterings, linear unmixing methods may not perform well in practical applications, and thus nonlinear unmixing has become an urgent problem to be solved. Usually, the mixing process in the observed scenarios is physically based, and many well-designed models have been proposed to interpret it. Recently, kernel-based nonlinear unmixing methods have been popularly studied to achieve a model-free and flexible representation of the nonlinearity. However, the existing kernel-based methods are mainly data-driven, which could make them fail to match the real physical mixing mechanism and result in the occurrence of overfitting. In this article, a kernel-based bilinear unmixing (KBU) method was proposed to transform the classic bilinear mixing models into their equivalent kernel forms that are more general and effective in expressing second-order scatterings. Two specific types of kernel transformations were designed, and the alternating direction method of multipliers (ADMM) was used to solve the kernel-transformed model-based optimization problem for unmixing. Moreover, the spatial prior was exploited to further improve the unmixing accuracy, and here we employ the total variation (TV) regularization as a paradigm. Experiments on synthetic data sets, physics-based simulated data sets, and real data were conducted to evaluate the algorithms. It is validated that our methods have better performance in abundance estimation and nonlinear reconstruction compared with other nonlinear unmixing methods. [ABSTRACT FROM AUTHOR]

Despite the successful applications of unsupervised sparse dimensionality reduction (USDR) in pattern recognition, the USDR still suffers from two challenges for hyperspectral images (HSIs), which limit its discriminative performance: first, it cannot be applied for dimensionality reduction using both training samples and testing samples; second, it lacks the ability to integrate the spectral with spatial information for improving the discriminative performance of HSIs. In order to tackle the first challenge, we extend it to a supervised scenario, which can be applied for both training samples and testing samples, namely dimensionality reduction sparse representation (DRSR). Then, we propose a novel method called local and global DRSR (LGDRSR) to integrate the spectral information and spatial information of HSIs to further improve the discriminative performance of HSIs. The proposed LGDRSR computes the distance information between pixels of HSIs including the whole samples and their corresponding locations with a unified metric matrix. Experimental results show the proposed LGDRSR outperforms other state-of-the-art algorithms significantly.

With the development of sensor technology, fusion of multiple remote sensors has aroused wide attention in the earth observation area. In this article, we propose to integrate the complementary information of hyperspectral image (HSI) and infrared image (IFI) based on mathematical morphological methods. HSI contains rich spectral information and spatial information, but the operation methods using only hyperspectral data are still subject to many restrictions. IFI can capture infrared rays radiated in the object, but it has no advantage in dealing with complex terrain classification. HSI and IFI can acquire different information of objects, and the information between these two kinds of data has great complementarity. In order to make full use of the information provided by HSI and IFI, this article proposes an HSI and IFI collaborative classification framework based on a Threshold-based Local Contain Profile (TLCP), where TLCP is our new design for suppressing interferes within spatial extractions. Specifically, the spatial information of HSI and IFI is extracted by TLCP, and then, these features are integrated and fed into the support vector machine for object classification. Experimentally, we compare the proposed method with the existing LCP and EP and evaluate the collaborative framework using GF5 satellite data collected over Hebei Province in China. Final results demonstrated the effectiveness of the proposed method.

Hyperspectral images (HSIs) showing objects belonging to several distinct target classes are characterized by dozens of spectral bands being available. However, some of these spectral bands are redundant and/or noisy, and hence, selecting highly informative and trustworthy bands for each class is a vital step for classification and for saving internal storage space; then the selected bands are termed a highly informative spectral band subset. We use a mutual information (MI)-based method to select the spectral band subset of a given class and two additional binary quantum classifiers, namely a quantum boost (Qboost) and a quantum boost plus (Qboost-Plus) classifier, to classify a two-label dataset characterized by the selected band subset. We pose both our MI-based band subset selection problem and the binary quantum classifiers as a quadratic unconstrained binary optimization (QUBO) problem. Such a quadratic problem is solvable with the help of conventional optimization techniques. However, the QUBO problem is an NP-hard global optimization problem, and hence, it is worthwhile for applying a quantum annealer. Thus, we adapted our MI-based optimization problem for selecting highly informative bands for each class of a given HSI to be run on a D-Wave quantum annealer. After the selection of these highly informative bands for each class, we employ our binary quantum classifiers to a two-label dataset on the D-Wave quantum annealer. In addition, we provide a novel multilabel classifier exploiting an error-encoding output code when using our binary quantum classifiers. As a real-world dataset in Earth observation, we used the well-known AVIRIS HSI of Indian Pine, north-western Indiana, USA. We can demonstrate that the MI-based band subset selection problem can be run on a D-Wave quantum annealer that selects the highly informative spectral band subset for each target class in the Indian Pine HSI. We can also prove that our binary quantum classifiers and our novel multilabel classifier generate a correct two- and multilabel dataset characterized by their selected bands and with high accuracy such as having been produced by conventional classifiers—and even better in some instances.