In recent years, deep-learning-based hyperspectral image (HSI) classification networks have become one of the most dominant implementations in HSI classification tasks. Among these networks, convolutional neural networks (CNNs) and attention-based networks have prevailed over other HSI classification networks. While convolutional neural networks with perceptual fields can effectively extract local features in the spatial dimension of HSI, they are poor at capturing the global and sequential features of spectral–spatial information; networks based on attention mechanisms, for example, Transformer, usually have better ability to capture global features, but are relatively weak in discriminating local features. This paper proposes a fusion network of convolution and Transformer for HSI classification, known as FusionNet, in which convolution and Transformer are fused in both serial and parallel mechanisms to achieve the full utilization of HSI features. Experimental results demonstrate that the proposed network has superior classification results compared to previous similar networks, and performs relatively well even on a small amount of training data.

Support vector machine (SVM) has a good effect in the supervised classification of hyperspectral images. In view of the shortcomings of the existing parallel structure SVM, this article proposes a non-parallel SVM model. Based on the traditional parallel boundary structure vector machine, this model adds an additional empirical risk minimization term to the original optimization problem by adding the least square term of the sample and obtains two non-parallel hyperplanes, respectively, forming a new non-parallel SVM algorithm to minimize the additional empirical risk of non-parallel SVM (Additional Empirical Risk Minimization Non-parallel Support Vector Machine, AERM-NPSVM). On the basis of AERM-NPSVM, the bias constraint is added to it, and AERM-NPSVM (BC-AERM-NPSVM) is further obtained. The experimental results show that, compared with the traditional parallel SVM model and the classical non-parallel SVM model, Twin Support Vector Machine (TWSVM), the new model, has a better effect in hyperspectral image classification and better generalization performance.

In this letter, a multiscale joint collaborative representation with locally adaptive dictionary (MLJCRC) method is proposed for hyperspectral image classification. Based on the joint collaborative representation model, instead of selecting only a single region scale, MLJCRC incorporates complementary contextual information into classification by multiplying different scales with distinct spatial structures and characteristics. Also, MLJCRC uses a locally adaptive dictionary to reduce the influence of irrelevant pixels on representation, which improves the classification accuracy. The results of experiments on Indian Pines data and Pavia University data demonstrate that the proposed method performs better than support vector machine, sparse representation classification, and other collaborative representation-based classifications.

Mapping forest disturbances using dense time series can timely identify disturbances at the subannual scale. However, these change detection methods using dense time series may be infeasible when not enough temporal observations are available. In this article, an online change detection algorithm that identifies forest disturbances at a subannual scale using spatial context from the sparse Landsat time series was proposed. First, the spatial normalized index that removed forest seasonality was prepared for establishing a simplified model instead of the harmonic model, thereby reducing the requirements for a high temporal frequency of clear observations for model initialization. Second, by using the spatial errors model to establish the simplified model, the normally distributed residual time series that removed the spatial autocorrelation were obtained. Third, the spatial statistic $t$ time series transformed from residual time series within a $3\times3$ spatial window were subsequently subjected to the exponentially weighted moving average $t$ chart (EWMA-t), which is a statistical process control chart for a short cycle corresponding to sparse Landsat time series. Fourth, disturbed pixels were labeled if the chart values persistently deviated from the control limits of the chart. The proposed algorithm was applied to a subtropical forest with low Landsat data availability and yielded an overall accuracy of 86% in the spatial domain and temporal accuracy of 93.7%, achieving accurate and timely identification of forest disturbances. The proposed method called the EWMA-t change detection (EWMATCD) algorithm provides an alternative for disturbance detection at the subannual scale in regions with low data availability.

In recent years, deep learning-based classification methods have been widely used for hyperspectral images (HSIs). However, in the existing transformer-based HSI classification methods, how to effectively and comprehensively utilize the rich information still have room for improvement, for example, when utilizing multiple image information, the comprehensive interaction between information have insufficient consideration. To address the above issues, cross-attention interaction, class token and patch token information, and multi-scale spatial information are addressed in a unified framework, and a cross-attention-based multi-information fusion transformer (CAMFT) for HSI classification was proposed, which includes the multi-scale patch embedding module, the residual connection based deepvit (RCD) module, and the double branch cross-attention (DBCA) module. First, the multi-scale patch embedding module is formed for multi-information preprocessing, accompanied by the built of different scale processing branches and the addition of learnable class tokens. Second, the RCD module is designed to utilize rich information from different layers, this module includes re-attention and residual connection. Third, a DBCA module is constructed to obtain more representative multi-information fusion features, this module not only integrates multi-scale patch information, but also effectively utilizes complementary information between class tokens and patch tokens in the interaction of two branches. Moreover, numerous experiments demonstrate that, compared with other state-of-the-art classification methods, the proposed CAMFT method achieves the optimal classification performance, especially with a small training sample size, it still has excellent performance.

In recent years, deep learning methods, especially convolutional neural networks (CNNs), have been gradually applied to the field of hyperspectral image (HIS) classification. Because the receptive fields of standard convolution are regular, fixed, and limited, CNNs usually only tend to focus on local formations, which cannot fully reflect the complex information in HSIs. To address the above issue, a novel HSI classification method based on pyramid feature extraction with deformable–dilated convolution (PD2C) is proposed. First, a pyramid feature extraction (PFE) model based on a multiscale double-branch module with deformable–dilated convolution (MDBD2) and a deformable downsampling module is proposed to extract local features. Second, transformer is used to extract global features. On this basis, complex information is well utilized for classification. Experiments on three public datasets show that the proposed PD2C method achieves optimal classification results compared with other state-of-the-art classification methods.