Multimodal land cover classification (LCC) of optical and SAR images has become a research hotspot. However, there are still two unsolved problems: the lack of a deep fusion mechanism and the neglect of the diversity of multimodal features. Inspired by ensemble learning, this letter proposes the cascaded multimodal forest-of-experts (CM2FEs) for deeper and broader fusion to further improve the performance of LCC. The proposed method first establishes the expert tree, then combines multiple trees at the same level into a forest, and finally forms a cascaded forest across different levels. Specifically, the novel designs include three points: 1) the multimodal expert tree is built based on linear projection and dynamic routing, with multiple layers of experts; it can acquire more discriminative multimodal features through deeper fusion; 2) the cascaded forest is formed by combining expert trees at the same level and different levels, which can effectively ensemble the knowledge learned by different trees; it can generate more diverse multimodal features through broader fusion; and 3) two expert exchange strategies are proposed to transfer knowledge between different trees and further optimize the feature fusion effect. Experiments show that the proposed method performs better than existing methods, and the mean IoU (mIoU) has been improved by at least 1.60%–3.25%.

The last decade has witnessed significant advances in semantic segmentation brought about by deep learning. However, existing methods only fit the data-label correspondence in a data-driven manner and do not fully conform to the abstraction and structuralization characteristics of the human visual cognition process, which limits the upper bounds of their performance. To this end, a multi-grained logical prototype (MGLP) method is proposed to rethink semantic segmentation based on these two key characteristics. Its novel design can be summarized as follows. 1) For abstraction, prototypes of the same class at different grain levels are established: a label generation method is proposed to automatically generate a multi-grained label space, which can guide the learning of the multi-grained prototypes for each class. 2) For structuralization, the intrinsic logical structure across different semantic levels is explicitly modeled: the horizontal metric relationships are established via metric relation operations on prototypes at the same grain level, to improve the discriminability between classes while taking the vertical semantic hierarchy into account. Moveover, the vertical logical relationships are established as the sub-to-super positive and super-to-sub negative constraints, to strengthen the semantic dependencies among prototypes at different grain levels. 3)MGLP is plug-and-play and can be directly combined with existing segmentation methods. Extensive experimental results indicate that MGLP can significantly improve the segmentation performance of existing methods, which opens up a new avenue for future research.

Recently, remote sensing images (RSIs) domain adaptation segmentation has been extensively studied. However, existing methods generally assume that source RSIs must be available, which is obviously an overly demanding condition and will increase unnecessary costs in practice. To this end, this letter takes the lead in exploring RSIs source-free adaptation segmentation, where only the offline model pretrained on the source domain and target RSIs are available. A novel method featuring prompt learning and vision foundation models is proposed, and the novelty design includes two aspects. First, to better adapt the general-purpose knowledge in the foundation model to different target RSIs, an attention-guided prompt tuning strategy is proposed, which can dynamically steer the knowledge at different layers and positions through prompts with different weights. Second, a feature alignment strategy with similarity distance is proposed for source-free domain adaptation by taking full advantage of the representation ability of the foundation model and the flexibility of prompt learning. Extensive experiments indicate that the performance of the proposed method is significantly superior to that of existing methods. Specifically, the mIoU of target RSIs has been improved by at least 3.14%~4.18%.

In the existing unsupervised domain adaptation (UDA) methods for remote sensing images (RSIs) semantic segmentation, class symmetry is a widely followed ideal assumption, where the source and target RSIs have exactly the same class space. In practice, however, it is often very difficult to find a source RSI with exactly the same classes as the target RSI. More commonly, there are multiple source RSIs available. And there is always an intersection or inclusion relationship between the class spaces of each source–target pair, which can be referred to as class asymmetry. Nevertheless, the class asymmetry domain adaptation segmentation of RSIs with multiple sources has not yet been explored. To this end, a novel class asymmetry RSIs domain adaptation method is proposed for the first time in this article, which consists of four key components. First, a multibranch segmentation network is built to learn an expert for each source RSI. Second, a novel collaborative learning method with the cross-domain mixing strategy is proposed, to supplement the class information for each source while achieving the domain adaptation of each source–target pair. Third, a pseudolabel generation strategy is proposed to effectively combine the strengths of different experts, which can be flexibly applied to two cases where the source class union is equal to or includes the target class set. Fourth, a multiview-enhanced knowledge integration module is developed for high-level knowledge routing and transfer from multiple domains to target predictions. The experimental results of six different class settings on airborne and spaceborne RSIs show that the proposed method can effectively perform the multisource domain adaptation in the case of class asymmetry, and the obtained segmentation performance of target RSIs is significantly better than the existing relevant methods.

In unsupervised domain adaptation (UDA) of remote sensing images (RSIs), the huge inter-domain discrepancies and intra-domain variances lead to complicated class-level relations. Specifically, the instances of the same class differ greatly while instances of different classes are similar, whether across different RSIs domains or within the same RSIs domain. However, existing methods cannot fully consider these problems, limiting the performance of UDA semantic segmentation of RSIs. To this end, this paper proposes a novel cross-domain multi-prototypes learning method, the core idea of which is to abstract the cross-and intra-domain class-level relations into multiple prototypes. Specifically, the multiple prototypes belonging to different classes can detailedly describe complex inter-class relations, and the multiple prototypes within the same class can better model rich intra-class relations. Further, the source and target samples are jointly used for prototypes calculation, to fully fuse the feature information of different RSIs. In a nutshell, utilizing the samples from different RSIs domains to learn multiple prototypes for each class can achieve better domain alignment at the class level. In addition, considering that RSIs simultaneously contain large targets with wide coverage and important small targets, two masked consistency learning strategies are designed to better explore the contextual structure of target RSIs and improve the quality of pseudo labels for prototype updating. The global consistency strategy can strengthen the utilization of global context relations, while the local consistency strategy can further improve the learning of local context details. Therefore, the proposed method is actually a prototype and context enhanced learning method for UDA semantic segmentation of RSIs. Extensive experiments demonstrate that the proposed method can achieve better performance than existing state-of-the-art UDA methods.

Abstract To address the existing problems of capsule networks in deep feature extraction and spatial‐spectral feature fusion of hyperspectral images, this paper proposes a hyperspectral image classification method that combines a deep residual 3D capsule network and Markov random field. Based on this method, the deep spatial‐spectral features of hyperspectral images are extracted using the deep residual 3D convolutional structure, the vector capsules of the features are obtained by the initial capsule layer and mapped into probability capsules via the 3D dynamic routing mechanism to construct the classification probability map, and the spatial structure of the classification results is regularised by the Markov random field to further improve the classification accuracy and performance of the images. Two sets of benchmark hyperspectral images, namely Indian Pines and Pavia University data sets, were used to conduct comparative experiments and ablation study. The experimental results showed that, compared with the conventional convolutional neural network and existing capsule network models, the proposed method not only improves the classification accuracy of the images but also partly eliminates the category noise and affords a more regular classification probability map.

In recent years, deep-learning-based hyperspectral image (HSI) classification methods have achieved significant development. The superior capability of feature extraction from these data-driven methods dramatically improves the classification performance. However, the previous methods usually require to retrain the network from scratch to obtain the capability of feature extraction adaptive for the target image when facing a new HSI to be classified, which is a time-consuming and redundant process. In this paper, we consider putting this process ahead and making the network have a robust capability of feature extraction with generalization through pre-training. Therefore, the network enables to directly extract features of the target HSI without re-training. For this purpose, we rethink the three-dimension (3D) HSI data from a perspective of spectral sequence, and we attempt to extract the spectral variation information as the spectrum motion feature. Then, we construct an unsupervised spectrum motion feature learning framework (SMF-UL), which can be pre-trained on mass unlabeled HSI data to learn the knowledge about perceiving spectral variation. Furthermore, to achieve the expansion of source data for pre-training, we develop an extendable training dataset construction method, which can integrate HSIs of different sizes, number of bands and sensors into a unified training set to utilize the rapidly growing mass unlabeled HSI data effectively. Finally, we use the trained network to directly extract the spectrum motion feature of the target HSI for classification, so the laborious re-training of the network can be avoided. Extensive experiments show that the proposed SMF-UL acquires the robust capability of feature extraction with generalization through unsupervised learning on mass unlabeled HSI data, and the classification performance of extracted spectrum motion feature is competitive to advanced in-domain and cross-domain methods, which shows its flexibility and superiority. The code of SMF-UL will be open at: https://github.com/sssssyf/SMF-UL.

The difficulties of obtaining sufficient labeled samples have always been one of the factors hindering deep learning models from obtaining high accuracy in hyperspectral image (HSI) classification. To reduce the dependence of deep learning models on training samples, meta learning methods have been introduced, effectively improving the classification accuracy in small sample set scenarios. However, the existing methods based on meta learning still need to construct a labeled source data set with several pre-collected HSIs, and must utilize a large number of labeled samples for meta-training, which is actually time-consuming and labor-intensive. To solve this problem, this paper proposes a novel unsupervised meta learning method with multiview constraints for HSI small sample set classification. Specifically, the proposed method first builds an unlabeled source data set using unlabeled HSIs. Then, multiple spatial-spectral multiview features of each unlabeled sample are generated to construct tasks for unsupervised meta learning. Finally, the designed residual relation network is used for meta-training and small sample set classification based on the voting strategy. Compared with existing supervised meta learning methods for HSI classification, our method can only utilize HSIs without any label for unsupervised meta learning, which significantly reduces the number of requisite labeled samples in the whole classification process. To verify the effectiveness of the proposed method, extensive experiments are carried out on 8 public HSIs in the cross-domain and in-domain classification scenarios. The statistical results demonstrate that, compared with existing supervised meta learning methods and other advanced classification models, the proposed method can achieve competitive or better classification performance in small sample set scenarios.

Recently, the field of hyperspectral image (HSI) classification is dominated by deep learning-based methods. However, training deep learning models usually needs a large number of labeled samples to optimize thousands of parameters. In this article, a deep multiview learning method is proposed to deal with the small sample problem of HSI. First, two views of an HSI scene are constructed by applying principal component analysis to different bands. Second, a deep residual network is designed to embed the different views of a sample to a latent space. The designed deep residual network is trained by maximizing agreement between differently augmented views of the same data sample via a contrastive loss in the latent space. Note that the training procedure of the designed deep residual network does not use labeled information. Therefore, the proposed method belongs to the category of unsupervised learning, which could alleviate the lack of labeled training samples. Finally, a conventional machine learning method (e.g., support vector machine) is used to complete the classification task in the learned latent space. To demonstrate the effectiveness of the proposed method, extensive experiments are carried on four widely used hyperspectral data sets. The experimental results demonstrate that the proposed method could improve the classification accuracy with small samples.