Convolutional neural network (CNN) is of great interest in machine learning and has demonstrated excellent performance hyperspectral image classification. In this paper, we propose a classification framework, called diverse region-based CNN, which can encode semantic context-aware representation to obtain promising features. With merging set discriminative appearance factors, the resulting CNN-based exhibits spatial-spectral context sensitivity that essential for accurate pixel The proposed...

Deep learning methods have been widely used in hyperspectral image classification and achieved state-of-the-art performance. Nonetheless, the existing deep are restricted by a limited receptive field, inflexibility, difficult generalization problems classification. To solve these problems, we propose HSI-BERT, where BERT stands for bidirectional encoder representations from transformers HSI imagery. The proposed HSI-BERT has global field that captures dependence among pixels regardless of...

Multisensor fusion is of great importance in Earth observation related applications. For instance, hyperspectral images (HSIs) provide wealthy spectral information while light detection and ranging (LiDAR) data elevation information, using HSI LiDAR together can achieve better classification performance. In this paper, an unsupervised feature extraction framework, named as patch-to-patch convolutional neural network (PToP CNN), proposed for collaborative data. More specific, a three-tower...

Domain adaptation techniques have been widely applied to the problem of cross-scene hyperspectral image (HSI) classification. Most existing methods use convolutional neural networks (CNNs) extract statistical features from data and often neglect potential topological structure information between different land cover classes. CNN-based approaches generally only model local spatial relationships samples, which largely limits their ability capture nonlocal relationship that would better...

Most domain adaptation (DA) methods in cross-scene hyperspectral image classification focus on cases where source data (SD) and target (TD) with the same classes are obtained by sensor. However, performance is significantly reduced when there new TD. In addition, alignment, as one of main approaches DA, carried out based local spatial information, rarely taking into account nonlocal information (nonlocal relationships) strong correspondence. A graph aggregation cross-domain few-shot learning...

Convolutional neural network (CNN) has been widely used in hyperspectral imagery (HSI) classification. Data augmentation is proven to be quite effective when training data size relatively small. In this letter, extensive comparison experiments are conducted with common methods, which draw an observation that methods can produce a limited and up-bounded performance. To address problem, new method, named as pixel-block pair (PBP), proposed greatly increase the number of samples. The method...

With the development of sensor technology, complementary data different sources can be easily obtained for various applications. Despite availability adequate multisource observation data, example, hyperspectral image (HSI) and light detection ranging (LiDAR) existing methods may lack effective processing on structural information transmission physical properties alignment, weakening ability multiple in collaborative classification task. The collaboration manner redundancy exclusion operator...

Joint use of multisensor information has attracted considerable attention in the remote sensing community. While applications land-cover observation benefit from diversity, integration technique is confronted with many challenges, including inconsistent size data, different data structures, uncorrelated physical properties, and scarcity training data. In this article, an fusion network, named interleaving perception convolutional neural network (IP-CNN), proposed for integrating...

The monitoring of coastal wetlands is great importance to the protection marine and terrestrial ecosystems. However, due complex environment, severe vegetation mixture, difficulty access, it impossible accurately classify identify their species with traditional classifiers. Despite integration multisource remote sensing data for performance enhancement, there are still challenges acquiring exploiting complementary merits from data. In this article, depthwise feature interaction network...

In this article, the intrinsic properties of hyperspectral imagery (HSI) are analyzed, and two principles for spectral-spatial feature extraction HSI built, including foundation pixel-level classification definition spatial information. Based on principles, scaled dot-product central attention (SDPCA) tailored is designed to extract information from a pixel (i.e., query be classified) pixels that similar an patch. Then, employed with HSI-tailored SDPCA module, network (CAN) proposed by...

Hyperspectral image (HSI) consists of abundant spectral and spatial characteristics, which contribute to a more accurate identification materials land covers. However, most existing methods hyperspectral analysis primarily focus on knowledge or coarse-grained information while neglecting the fine-grained morphological structures. In classification task complex objects, differences can help search for boundary classes, e.g., forestry tree species. Focusing subtle traits extraction,...

Due to the limitations of single-source data, joint classification using multisource remote sensing data has received increasing attention. However, existing methods still have certain shortcomings when faced with feature extraction from and fusion between data. In this article, a method based on multiscale interactive information (MIFNet) for hyperspectral synthetic aperture radar (SAR) image is proposed. First, (MIIE) block designed extract meaningful information. Compared traditional...

Joint classification using multisource remote sensing data for Earth observation is promising but challenging. Due to the gap of imaging mechanism and imbalanced information between data, integrating complementary merits interpretation still full difficulties. In this article, a method based on asymmetric feature fusion, named fusion network (AsyFFNet), proposed. First, weight-share residual blocks are utilized extraction while keeping separate batch normalization (BN) layers. training...

In recent years, remote sensing scene classification is one of research hotspots and has played an important role in the field intelligent interpretation data. However, various complex objects backgrounds form a variety scenes through spatial combination correlation, which brings great challenges to accurately classify different scenes. Among them, insufficient feature difference brought about unbalanced change background target between inter-class sample representation inconsistency caused...

Due to rich spectral and spatial information, the combination of hyperspectral multispectral images (MSIs) has been widely used for Earth observation, such as wetland classification. However, mining meaningful features effective fusion multisource remote sensing data are still urgent problems be solved. In this article, graph-feature-enhanced selective assignment network (GSANet) is proposed. On one hand, a graph feature extraction module (GFEM) designed extract topological structure...

With the recent development of joint classification hyperspectral image (HSI) and light detection ranging (LiDAR) data, deep learning methods have achieved promising performance owing to their locally sematic feature extracting ability. Nonetheless, limited receptive field restricted convolutional neural networks (CNNs) represent global contextual sequential attributes, while visual transformers (VITs) lose local semantic information. Focusing on these issues, we propose a fractional Fourier...

Text information including extensive prior knowledge about land cover classes has been ignored in hyperspectral image classification (HSI) tasks. It is necessary to explore the effectiveness of linguistic mode assisting HSI classification. In addition, large-scale pre-training image-text foundation models have demonstrated great performance a variety downstream applications, zero-shot transfer. However, most domain generalization methods never addressed mining modal improve model. To...

Domain adaption (DA) is a challenging task that integrates knowledge from source domain (SD) to perform data analysis for target domain. Most of the existing DA approaches only focus on single-source-single-target setting. In contrast, multisource (MS) collaborative utilization has been extensively used in various applications, while how integrate with MS collaboration still faces great challenges. this article, we propose multilevel network (MDA-NET) promoting information and cross-scene...

Hyperspectral and multispectral images (HS/MS) fusion classification as an important branch of data quality improvement interpretation, has attracted increasing attention in recent years. However, the unavailable sensor prior still limits performance many traditional methods, consequently deteriorating results. Despite unsupervised methods based on convolutional neural network (CNN) making a lot attempts to mitigate limitations, challenges with extracting long-range dependencies hamper...

Convolutional neural networks (CNN) have attracted increasing attention in the field of multimodal cooperation. Recently, adoption CNN-based methods has achieved remarkable performance multisource remote sensing data classification. However, it is still confronted with challenges aspect complementarity extraction. In this paper, adversarial complementary learning strategy embedded into CNN model called ACL-CNN, which employed to extract information data. The proposed ACL-CNN able filter out...

Advances in multisource remote sensing have allowed for the development of more comprehensive observation. The adoption deep convolutional neural networks (CNN) naturally includes spatial-spectral information, which has achieved promising performance data classification. However, challenges are still found with extraction spatial distribution and spectrum relationships, eventually limit classification performance. To solve issue, a perception network (S2PNet) is proposed to extract...

To solve the problem of supervised convolutional neural network (CNN) models suffering from limited samples, a two-channel CNN is developed for medical hyperspectral images (MHSI) classification tasks. In proposed network, one channel end-to-end denoted as EtoE-Net, designed to realize unsupervised learning, obtaining representative and global fused features with fewer noises, by building pixel-by-pixel mapping between two source data, i.e., original MHSI data its principal component. On...

Remote sensing scene understanding is a highly challenging task, and has gradually emerged as research hotspot in the field of intelligent interpretation remote data. Recently, use convolutional neural networks (CNNs) been proven to be fruitful advancement. However, with emergence visual transformers (ViTs), limitations traditional small kernels directly capturing large receptive have posed significant challenges their dominant role. Additionally, fixed neuron connections between different...

Mixup-based data augmentation has been proven to be beneficial the regularization of models during training, especially in remote-sensing field where training is scarce. However, process augmentation, methods ignore target proportion different inputs and keep linear insertion ratio consistent, which leads response label space even if no effective objects are introduced mixed image due randomness process. Moreover, although some previous works have attempted utilize multimodal interaction...