Convolutional neural networks (CNNs) have been widely used in hyperspectral image (HSI) classification tasks because of their excellent local spatial feature extraction capabilities. However, because it is difficult to establish dependencies between long sequences of data for CNNs, there are limitations in the process of processing hyperspectral spectral sequence features. To overcome these limitations, inspired by the Transformer model, a spatial–spectral transformer with cross-attention (CASST) method is proposed. Overall, the method consists of a dual-branch structures, i.e., spatial and spectral sequence branches. The former is used to capture fine-grained spatial information of HSI, and the latter is adopted to extract the spectral features and establish interdependencies between spectral sequences. Specifically, to enhance the consistency among features and relieve computational burden, we design a spatial–spectral cross-attention module with weighted sharing to extract the interactive spatial–spectral fusion feature intra Transformer block, while also developing a spatial–spectral weighted sharing mechanism to capture the robust semantic feature inter Transformer block. Performance evaluation experiments are conducted on three hyperspectral classification datasets, demonstrating that the CASST method achieves better accuracy than the state-of-the-art Transformer classification models and mainstream classification networks.

With the increasing spectral dimension of hyperspectral images (HSI), how correctly choose bands based on band correlation and information has become more significant, but also complicated. Band selection is a combinatorial optimization problem, and intelligent optimization algorithms have been shown to be crucial in solving combinatorial optimization problems. However, major of them only use a single objective as the selection index, while neglecting the overall features of hyperspectral images, which may lead to inaccuracy in object detection. To tackle this, we propose a band selection method based on a multi-objective cuckoo search algorithm (MOCS) when constructing a multi-objective unsupervised band selection model based on the amount of information and correlation of the bands (MOCS-BS). Specifically, an adaptive strategy based on population crowding degree is first proposed to assist Lévy flight in overcoming the influence of the parameter constancy. Then, an information-sharing strategy based on grouping and crossover is designed to balance the search ability between global exploration and local exploitation, which can overcome the shortcomings caused by the lack of information interaction between individuals. Finally, the HSI classification experiments are performed by Random Forest and KNN classifiers based on the subset of bands selected by the proposed MOCS-BS method. The proposed method is compared with state-of-the-art algorithms including neighborhood grouping normalized matched filter (NGNMF) and multi-objective artificial bee colony with band selection (MABC-BS) on four HSI datasets. The experimental results demonstrate that MOCS-BS is more effective and robust than other methods.

Hyperspectral images (HSIs) have high spatial resolution and spectral resolution, and using HSI as a change detection (CD) data source is crucial for detecting surface changes. However, there is a large amount of real noise in HSIs, and most deep-learning-based CD methods require a large number of ground-truth labels for training, which is difficult and expensive to label manually. To reduce the dependence of CD on ground-truth labels and weaken the interference of noise on CD in HSIs, in this article, we propose an HSI change detection framework with a self-supervised contrastive learning pretrained model (CDSCL). CDSCL consists of two parts: self-supervised contrastive learning pretrained model and CD classification network. The main contributions of this article are as follows: a data augmentation strategy based on Gaussian noise is proposed to improve the ability of the model to extract variation information from HSIs with different random Gaussian noises; based on the information bottleneck theory, a progressive feature extraction module is developed to remove redundant or irrelevant details in changing information spectrum; and a contrastive loss function based on the Pearson correlation coefficient and negative cosine correlation is designed to make the features extracted by the two branches of the siamese network close to each other. Experimental results on four real hyperspectral datasets demonstrate that the CD performance of CDSCL outperforms the most representative CD methods.

Hyperspectral image (HSI) characteristics of the abundant spectral information are favores by many scholars, but the challenge is how to select relevant features from such high dimensional data. Band selection (BS), one of the most fundamental dimensionality reduction (DR) techniques, removes redundant bands while providing a subset of bands that can preserve high information content and low noise for further HSI classification. Cuckoo search algorithm (CS) is well-known for its high performance of searching relevant features but struggles to get rid of local extremes in the late iteration. Therefore, in this paper, an unsupervised band selection method based on the heterogeneous cuckoo search algorithm with matched filter (HCS-MF) is proposed for hyperspectral image classification, in which an optimization model is constructed based on the sensitivity of the matching filter to noise. To reduce the similarity between selected bands, a mapping method based on neighborhood band grouping is proposed. In addition, an automatic recommendation strategy based on sliding spectrum decomposition (SSD) is proposed to determine the minimum recommended number of selected bands in different scenes. The superiority of the selected subset of bands is verified by Random Forest, Support Vector Machine (SVM) and Edge-Preserving Filtering-based SVM (EPF-SVM) classifiers. Experimental results on three well-known datasets demonstrate the robustness and superiority of the proposed HCS-MF algorithm compared with the state-of-the-art methods, such as marginalized graph self-representation (MGSR), neighborhood grouping normalized matched filter (NGNMF), and region-wise multiple graph fusion (RMGF).

Background subtraction is one of the most fundamental and challenging tasks in computer vision. Many background subtraction algorithms work well under the assumption that backgrounds are static over short time periods, but degrade dramatically in dynamic scenes such as swaying trees, rippling water, and waving curtains. In this paper, we propose an effective background subtraction method to address these difficulties by combining color features with texture features in the ViBe framework. Specifically, we present a novel Local Compact Binary Count (LCBC) feature, which can capture local binary gray-scale difference information, and totally discard the local binary structural information. The effective fusion of color and LCBC information significantly improves the performance of the ViBe model, making it very robust to background variations while still highlighting moving objects. We further embed the Total Variation (TV) norm regularization technique into the proposed method, which can enhance the spatial smoothness of foreground objects, thereby further improving the accuracy of the method. We evaluate the proposed method against ten sequences containing dynamic backgrounds and show that our method outperforms many state-of-the-art methods in reducing false positives without compromising reasonable foreground definitions. Experimental results on challenging well-known data sets demonstrate that the proposed method works effectively on a wide range of dynamic background scenes.