Search

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.

Hyperspectral images (HSIs) are used in a large number of real-world applications. HSI classification (HSIC) is a challenging task due to high interclass similarity, high intraclass variability, overlapping, and nested regions. The 2-D convolutional neural network (CNN) is a viable classification approach since HSIC depends on both spectral-spatial information. The 3-D CNN is a good alternative for improving the accuracy of HSIC, but it can be computationally intensive due to the volume and spectral dimensions of HSI. Furthermore, these models may fail to extract quality feature maps and underperform over the regions having similar textures. This work proposes a 3-D CNN model that utilizes both spatial-spectral feature maps to improve the performance of HSIC. For this purpose, the HSI cube is first divided into small overlapping 3-D patches, which are processed to generate 3-D feature maps using a 3-D kernel function over multiple contiguous bands of the spectral information in a computationally efficient way. In brief, our end-to-end trained model requires fewer parameters to significantly reduce the convergence time while providing better accuracy than existing models. The results are further compared with several state-of-the-art 2-D/3-D CNN models, demonstrating remarkable performance both in terms of accuracy and computational time.

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.

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.

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.

Change detection is a popular topic in remote sensing that is generally constrained to two remote sensing images captured at two different times. However, the optimal type of remote sensing image for change detection tasks has not yet been determined. The use of only hyperspectral images (HSIs) with low spatial resolution or multispectral images (MSIs) with low spectral resolution cannot obtain satisfactory change detection results. In this article, we propose the fusion of simultaneously captured low spatial resolution HSIs and low spectral resolution MSIs with the use of a tensor regression-based method to detect change regions from the fused images at two different time points. In this method, nonlocal couple tensor CP decomposition is initially applied to fuse the HSIs and MSIs. A difference image is then obtained by subtracting the fused images at two different time points. Thereafter, the tensors are extracted from the difference image and the tensor regression-based method is used to classify the difference image and detect the final change results. Experimental results from three real datasets suggest that the proposed method substantially outperforms the existing state-of-the-art change detection methods as well as any change detection methods using single-source images.

To reconstruct images with high spatial resolution and high spectral resolution, one of the most common methods is to fuse a low-resolution hyperspectral image (HSI) with a high-resolution (HR) multispectral image (MSI) of the same scene. Deep learning has been widely applied in the field of HSI-MSI fusion, which is limited with hardware. In order to break the limits, we construct an unsupervised multiattention-guided network named UMAG-Net without training data to better accomplish HSI-MSI fusion. UMAG-Net first extracts deep multiscale features of MSI by using a multiattention encoding network. Then, a loss function containing a pair of HSI and MSI is used to iteratively update parameters of UMAG-Net and learn prior knowledge of the fused image. Finally, a multiscale feature-guided network is constructed to generate an HR-HSI. The experimental results show the visual and quantitative superiority of the proposed method compared to other methods.

Recently, many spectral-special classification models have emerged one after another in the remote sensing community. These models aim to introduce the spatial information of the pixel to improve the accuracy of the class attribute of the pixel. However, for the spectral-spatial classification algorithms, not all pixels need to introduce the corresponding spatial information since the use of a large amount of spatial information has a costly time. To solve this problem, this paper proposes a robust dual-stage spatial embedding (RDSSE) model for spectral-spatial classification of hyperspectral image, which is composed of the following several main steps: First, an over-segmentation algorithm is employed to cluster original hyperspectral image into many superpixel blocks with shape adaptive characteristics. Then, we design a $k$ -peak criterion to fuse the spectral feature of pixels within and between superpixels. Next, a low time-consumption spectral classifier is introduced to conduct primary classification for a testing pixel to achieve the corresponding probability distribution. Specifically, the difference between the probability of the largest class and that of the second-largest class is served as class confidence. Finally, the predicted label of the low-confidence testing pixels is reclassified based on a high-accuracy spectral-spatial classification method. Experimental results on several real images illustrated that the proposed RDSSE method can obtain superior performance with respect to several competitive methods.

Convolutional neural networks (CNNs) have becomequite popular for solving many different tasks in remote sensingdata processing. The convolution is a linear operation whichextracts features from the input data. However, nonlinear operationsare able to better characterize the internal relationshipsand hidden patterns within complex remote sensing data, suchas hyperspectral images (HSIs). Morphological operations arepowerful nonlinear transformations for feature extraction thatpreserve the essential characteristics of the image, such asborders, shape and structural information. In this paper, anew end-to-end morphological deep learning framework (calledMorphConvHyperNet) is introduced. The proposed approachefficiently models nonlinear information during the trainingprocess of HSI classification. Specifically our method includesspectral and spatial morphological blocks to extract relevantfeatures from the HSI input data. These morphological blocksconsist of two basic 2D morphological operators (erosion anddilation) in the respective layers, followed by a weighted combinationof the feature maps. Both layers can successfully encodethe nonlinear information related to shape and size, playing animportant role in classification performance. Our experimentalresults, obtained on five widely used HSIs, reveal that our newlyproposed MorphConvHyperNet offers comparable (and evensuperior) performance when compared to traditional 2D and 3DCNNs for HSI classification.

Attribute profiles are well-acknowledged as one of the most significant techniques to characterize spectral-spatial properties of a hyperspectral image. The spectral-spatial content of an attribute profile is influenced by the threshold values considered during its construction. In this article, we propose a robust method to detect the threshold values automatically by overcoming the limitations of the existing techniques. The proposed method employs a tree structure representing the connected components of the image and evaluates attribute values at each node. Then, a total characteristic function (TCF) is defined that represents these attribute values in a nondecreasing order. The defined TCF is analyzed using a novel technique to detect a few informative thresholds for the construction of a low-dimensional attribute profile representing substantial spectral-spatial information. The proposed threshold detection method is computationally efficient. To assess the effectiveness of the proposed technique experiments are conducted on three real hyperspectral datasets using six different attributes and the results are compared to the recent state-of-the-art method. The results demonstrate that the proposed method has several advantages over the existing state-of-the-art method.

Morphological features have shown promising performances for hyperspectral image (HSIs) classification, as they can efficiently extract the multilevel spatial information of HSIs. However, the objects in the scenes are always with different sizes and shapes, making it difficult to excavate spatial information of important structures completely by the existing morphological methods. To address this problem, we propose a novel two-stage framework based on morphology and superpixel. Specifically, we propose self-dual saliency profiles (SPs) based on a saliency measure considering the grayscale contrast within objects and edge information. SPs are hierarchical features that characterize spatial information for salient objects whose saliency index is the significant local maxima. The SPs are constructed based on a two-step algorithm. First, all salient objects of different shapes in the scene are preserved, and the undesired spatial details are discarded by attribute filters based on the saliency measure. Second, the morphological feature is generated based on the organization structure of salient objects in the scene, which provides hierarchical spatial features of the image. Then, the superpixel segmentation is performed on each of the extracted SPs on the basis of the spatial information of salient objects. And, two types of superpixel-based features are extracted from SPs to exploit the information in SPs. The extracted innersuperpixel and intersuperpixel features are fused with spectral information to produce the classification map. The experiments conducted on three HSIs show that, the proposed approach significantly outperforms the other state-of-the-art methods.

Hyperspectral compressed sensing (HCS) is a new imaging method that effectively reduces the power consumption of data acquisition. In this article, we present a novel HCS algorithm by incorporating spatial–spectral hybrid compressed sensing, followed by a reconstruction based on spectral unmixing. At the sampling stage, the measurements are acquired by a spatial–spectral hybrid compressive sampling scheme to preserve the necessary information for the following spectral unmixing, where spatial compressive sampling mainly retains the endmember information, and spectral compressive sampling mainly retains the abundance information. Due to the limitations of the traditional linear mixed model, an improved mixed model is proposed for HCS reconstruction, which considers spectral variability, nonlinear mixing, and other factors. At the reconstruction stage, based on the improved mixed model, semi-nonnegative matrix factorization is introduced to realize spectral unmixing on the measurements to achieve the final reconstruction by using an alternate iteration manner. The proposed algorithm is tested on real hyperspectral data, and the selection of parameters is fully analyzed. Experimental results demonstrate that the proposed algorithm can significantly outperform state-of-the-art HCS algorithms in terms of reconstruction performance.

Generative adversarial networks (GANs) have shown striking performances in computer vision applications to augment virtual training samples (VTS). However, the VTS generating by GANs in the context of hyperspectral image classification suffer from structural inconsistency due to the insufficient number of training samples in order to learn high-order features from the discriminator. This work addresses the scarcity of training samples by designing a GAN, in which the performance of discriminator is improved to produce more structurally coherent VTS. In the proposed method, by splitting the discriminator into two parts, GAN undertakes two tasks: the main task is to learn to distinguish between real and fake samples, and the auxiliary task is to learn to distinguish structurally corrupted and real samples. With this setup, GAN will produce real-like VTS with a higher variation than conventional GAN. Furthermore, in order to reduce the computational cost, subspace-based dimension reduction was performed to obtain the dominant features around the training samples to generate meaningful patterns from the original ones to be used in the learning phase. Based on the experimental results on real, and well-known hyperspectral benchmark images, the proposed method improves the performance compared with GANs-related, and conventional data augmentation strategies1.

Clustering algorithms play an essential and unique role in classification tasks, especially when annotated data are unavailable or very scarce. Current clustering approaches in remote sensing are mostly designed for a single data source, such as hyperspectral image (HSI), while, nowadays, multisensor data are being routinely acquired. In this article, we propose a multiview subspace clustering model that exploits effectively the rich information from multiple features extracted either from a single data source (HSI) or from multiple sources that we call generically multiviews of the same scene. An important novelty of our approach is that it integrates local and nonlocal spatial information from each view in a unified framework. Our model learns a common intrinsic cluster structure from view-specific subspace representations by a new decomposition-based scheme. In addition, we develop innovative manifold-based spatial regularization as a hybrid hypergraph, which merges local and nonlocal spatial context and improves, thereby, the learning of view-specific structures. We develop an efficient algorithm to solve the resulting optimization problem. Extensive experiments on real data sets demonstrate the superior clustering performance over the state of the art.

Nowadays, a large number of remote sensing instruments are providing a massive amount of data within the frame of different Earth Observation missions. These instruments are characterized by the wide variety of data they can collect, as well as the impressive volume of data and the speed at which it is acquired. In this sense, hyperspectral imaging data has certain properties that make it difficult to process, such as its large spectral dimension coupled with problematic data variability. To overcome these challenges, convolutional neural networks have been proposed as classification models because of their ability to extract relevant spectral–spatial features and learn hidden patterns, along their great architectural flexibility. Their high performance relies on the convolution kernels to exploit the spatial relationships. Thus, filter design is crucial for the correct performance of models. Nevertheless, hyperspectral data may contain objects with different shapes and orientations, preventing filters from “seeing everything possible” during the decision making. To overcome this limitation, this paper proposes a novel adaptable convolution model based on deforming kernels combined with deforming convolution layers to fit their effective receptive field to the input data. The proposed adaptable convolutional network (named DKDCNet) has been evaluated over two well-known hyperspectral scenes, demonstrating that it is able to achieve better results than traditional strategies with similar computational cost for HSI classification.

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.

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.

Many studies analyzing spaceborne hyperspectral images (HSIs) have so far struggled to deal with a lack of pure pixels due to complex mixtures of urban surface materials. Recently, an alternative concept of gradients in urban surface material composition has been proposed and successfully applied to map cities with spaceborne HSIs without the requirement for a previous determination of pure pixels. The gradient concept treats all pixels as mixed and aims to describe and quantify gradual transitions in the cover fractions of surface materials. This concept presents a promising approach to tackle urban mapping using spaceborne HSIs. However, since gradients are determined in a data-driven way, their transferability within urban areas needs to be investigated. For this purpose, we analyze the robustness of urban surface material gradients and their dependence across six systematic and three simple random sampling schemes. The results show high similarity between nine sampling schemes in the primary gradient feature space (Pspace) and individual gradient feature spaces (Ispaces). Comparing the Pspace with the Ispaces, the Mantel statistics show the resemblance of samples' distribution in the Pspace, and each Ispace is rather strong with high credibility, as the significance level is P < 0.01. Therefore, it can be concluded that the material gradients defined in the test area are independent of the specific sampling scheme. This study paves the way for subsequent analysis of the stability of urban surface material gradients and the interpretation of material gradients in other urban environments.

The storage and processing of remotely sensed hyperspectral images (HSIs) is facing important challenges due to the computational requirements involved in the analysis of these images, characterized by continuous and narrow spectral channels. Although HSIs offer many opportunities for accurately modeling and mapping the surface of the Earth in a wide range of applications, they comprise massive data cubes. These huge amounts of data impose important requirements from the storage and processing points of view. The support vector machine (SVM) has been one of the most powerful machine learning classifiers, able to process HSI data without applying previous feature extraction steps, exhibiting a robust behaviour with high dimensional data and obtaining high classification accuracies. Nevertheless, the training and prediction stages of this supervised classifier are very time-consuming, especially for large and complex problems that require an intensive use of memory and computational resources. This paper develops a new, highly efficient implementation of SVMs that exploits the high computational power of graphics processing units (GPUs) to reduce the execution time by massively parallelizing the operations of the algorithm while performing efficient memory management during data-reading and writing instructions. Our experiments, conducted over different HSI benchmarks, demonstrate the efficiency of our GPU implementation.

The convolutional neural network (CNN) has been gradually applied to the hyperspectral images (HSIs) classification, but the lack of training samples caused by the difficulty of HSIs sample marking and ignoring of correlation between spatial and spectral information seriously restrict the HSIs classification accuracy. In an attempt to solve these problems, this paper proposes a dual-branch extraction and classification method under limited samples of hyperspectral images based on deep learning (DBECM). At first, a sample augmentation method based on local and global constraints in this model is designed to augment the limited training samples and balance the number of different class samples. Then spatial-spectral features are simultaneously extracted by the dual-branch spatial-spectral feature extraction method, which improves the utilization of HSIs data information. Finally, the extracted spatial-spectral feature fusion and classification are integrated into a unified network. The experimental results of two typical datasets show that the DBECM proposed in this paper has certain competitive advantages in classification accuracy compared with other public HSIs classification methods, especially in the Indian pines dataset. The parameters of the overall accuracy (OA), average accuracy (AA), and Kappa of the method proposed in this paper are at least 4.7%, 5.7%, and 5% higher than the existing methods.

The applications of extreme learning machine (ELM) to the hyperspectral-image (HSI) classification have attracted a great deal of research attention because of its excellent performance and fast learning speed. However, conventional ELM is unable to achieve satisfactory accuracy since it only exploits the spectral information to conduct the HSI classification. To address the above issues, we propose a novel classification algorithm based on both spectral and multiscale spatial information, referred to as ELM with enhanced composite feature (ELM-ECF). To be specific, we adopt the original ELM, exploit a multiscale spatial weighted-mean-filtering-based approach to extract multiple spatial information, and use the majority vote method to select the final classification result. The proposed ELM-ECF significantly improves the classification accuracy of the original ELM. Experimental results on three public HSIs (i.e., Indian Pines data set, Pavia University data set, and Salinas data set) illustrate that the proposed ELM-ECF outperforms a variety of the state-of-the-art HSI classification counterparts in terms of classification accuracy.

Generalized bilinear model (GBM) has received extensive attention in the field of hyperspectral nonlinear unmixing. Traditional GBM unmixing methods are usually assumed to be degraded only by additive white Gaussian noise (AWGN), and the intensity of AWGN in each band of hyperspectral image (HSI) is assumed to be the same. However, the real HSIs are usually degraded by mixture of various kinds of noise, which include Gaussian noise, impulse noise, dead pixels or lines, stripes, and so on. Besides, the intensity of AWGN is usually different for each band of HSI. To address the above mentioned issues, we propose a novel nonlinear unmixing method based on the bandwise generalized bilinear model (NU-BGBM), which can be adapted to the presence of complex mixed noise in real HSI. Besides, the alternative direction method of multipliers (ADMM) is adopted to solve the proposed NU-BGBM. Finally, extensive experiments are conducted to demonstrate the effectiveness of the proposed NU-BGBM compared with some other state-of-the-art unmixing methods.

An extinction profile (EP) is an effective spatial–spectral feature extraction method for hyperspectral images (HSIs), which has recently drawn much attention. However, the existing methods utilize the EPs in a stacking way, which is hard to fully explore the information in EPs for HSI classification. In this paper, a novel fusion framework termed EPs-fusion (EPs-F) is proposed to exploit the information within and among EPs for HSI classification. In general, EPs-F includes the following two stages. In the first stage, by extracting the EPs from three independent components of an HSI, three complementary groups of EPs can be constructed. For each EP, an adaptive superpixel-based composite kernel strategy is proposed to explore the spatial information within an EP. The weights to create the composite kernel and the number of superpixels are automatically determined based on the spatial information of each EP. In the second stage, since the different EPs contain highly complementary information, a simple yet effective decision fusion method is further applied to obtain the final classification result. Experiments on three real HSI data sets verify the qualitative and quantitative superiority of the proposed EPs-F method over several state-of-the-art HSI classifiers.

Sea ice is one of the causes of marine disasters. The classification of sea ice images is an important part of sea ice detection. The labeled samples in hyperspectral sea ice image classification are difficult to acquire, which causes minor sample problems. In addition, most of the current sea ice classification methods mainly use spectral features for shallow learning, which also limits further improvement of the sea ice classification accuracy. Therefore, this paper proposes a hyperspectral sea ice image classification method based on the spectral-spatial-joint feature with deep learning. The proposed method first extracts sea ice texture information by the gray-level co-occurrence matrix (GLCM). Then, it performs dimensionality reduction and a correlation analysis of the spectral information and spatial information of the unlabeled samples, respectively. It eliminates redundant information by extracting the spectral-spatial information of the neighboring unlabeled samples of the labeled sample and integrating the information with the spectral and texture data of the labeled sample to further enhance the quality of the labeled sample. Lastly, the three-dimensional convolutional neural network (3D-CNN) model is designed to extract the deep spectral-spatial features of sea ice. The proposed method combines relevant textural features and performs spectral-spatial feature extraction based on the 3D-CNN model by using a large amount of unlabeled sample information. In order to verify the effectiveness of the proposed method, sea ice classification experiments are carried out on two hyperspectral data sets: Baffin Bay and Bohai Bay. Compared with the CNN algorithm based on a single feature (spectral or spatial) and other CNN algorithms based on spectral-spatial features, the experimental results show that the proposed method achieves better sea ice classification (98.52% and 97.91%) with small samples. Therefore, it is more suitable for classifying hyperspectral sea ice images.

Deep neural networks (DNNs) have emerged as a relevant tool for the classification of remotely sensed hyperspectral images (HSIs), with convolutional neural networks (CNNs) being the current state-of-the-art in many classification tasks. However, deep CNNs present several limitations in the context of HSI supervised classification. Although deep models are able to extract better and more abstract features, the number of parameters that must be fine-tuned requires a large amount of training data (using small learning rates) in order to avoid the overfitting and vanishing gradient problems. The acquisition of labeled data is expensive and time-consuming, and small learning rates forces the gradient descent to use many small steps to converge, slowing down the runtime of the model. To mitigate these issues, this paper introduces a new deep CNN framework for spectral-spatial classification of HSIs. Our newly proposed framework introduces shortcut connections between layers, in which the feature maps of inferior layers are used as inputs of the current layer, feeding its own output to the rest of the the upper layers. This leads to the combination of various spectral-spatial features across layers that allows us to enhance the generalization ability of the network with HSIs. Our experimental results with four well-known HSI datasets reveal that the proposed deep&dense CNN model is able to provide competitive advantages in terms of classification accuracy when compared to other state-of-the-methods for HSI classification.

In this paper, we introduce a novel classification framework for hyperspectral images (HSIs) by jointly employing spectral, spatial, and hierarchical structure information. In this framework, the three types of information are integrated into the SVM classifier in a way of multiple kernels. Specifically, the spectral kernel is constructed through each pixel’s vector value in the original HSI, and the spatial kernel is modeled by using the extended morphological profile method due to its simplicity and effectiveness. To accurately characterize hierarchical structure features, the techniques of Fish-Markov selector (FMS), marker-based hierarchical segmentation (MHSEG) and algebraic multigrid (AMG) are combined. First, the FMS algorithm is used on the original HSI for feature selection to produce its spectral subset. Then, the multigrid structure of this subset is constructed using the AMG method. Subsequently, the MHSEG algorithm is exploited to obtain a hierarchy consist of a series of segmentation maps. Finally, the hierarchical structure information is represented by using these segmentation maps. The main contributions of this work is to present an effective composite kernel for HSI classification by utilizing spatial structure information in multiple scales. Experiments were conducted on two hyperspectral remote sensing images to validate that the proposed framework can achieve better classification results than several popular kernel-based classification methods in terms of both qualitative and quantitative analysis. Specifically, the proposed classification framework can achieve 13.46–15.61% in average higher than the standard SVM classifier under different training sets in the terms of overall accuracy.