Despite success in the same-scene hyperspectral image classification (HSIC), for the cross-scene classification, samples between source and target scenes are not drawn from the independent and identical distribution, resulting in significant performance degradation. To tackle this issue, a novel unsupervised domain adaptation (UDA) framework towards multi-level features and decision boundaries (ToMF-B) is proposed for the cross-scene HSIC, which can align task-related features and learn task-specific decision boundaries in parallel. Based on the maximum classifier discrepancy, a two-stage alignment scheme is proposed to bridge the interdomain gap and generate discriminative decision boundaries. In addition, to fully learn task-related and domain-confusing features, a CNN and Transformer-based multi-level features extractor (generator) is developed to enrich the feature representation of two domains. Furthermore, to alleviate the harm even the negative transfer to UDA caused by task-irrelevant features, a task-oriented feature decomposition method is leveraged to enhance the task-related features while suppressing task-irrelevant features, enabling the aligned domain-invariant features can be contributed to the classification task explicitly. Extensive experiments on three cross-scene HSI benchmarks have validated the effectiveness of the proposed framework.

Deep learning (DL) based methods represented by convolutional neural networks (CNNs) are widely used in hyperspectral image classification (HSIC). Some of these methods have strong ability to extract local information, but the extraction of long-range features is slightly inefficient, while others are just the opposite. For example, limited by the receptive fields, CNN is difficult to capture the contextual spectral-spatial features from a long-range spectral-spatial relationship. Besides, the success of DL-based methods is greatly attributed to numerous labeled samples, whose acquisition are time-consuming and cost-consuming. To resolve these problems, a hyperspectral classification framework based on multi-attention Transformer (MAT) and adaptive superpixel segmentation-based active learning (MAT-ASSAL) is proposed, which successfully achieves excellent classification performance, especially under the condition of small-size samples. Firstly, a multi-attention Transformer network is built for HSIC. Specifically, the self-attention module of Transformer is applied to model long-range contextual dependency between spectral-spatial embedding. Moreover, in order to capture local features, an outlook-attention module which can efficiently encode fine-level features and contexts into tokens is utilized to improve the correlation between the center spectral-spatial embedding and its surroundings. Secondly, aiming to train a excellent MAT model through limited labeled samples, a novel active learning (AL) based on superpixel segmentation is proposed to select important samples for MAT. Finally, to better integrate local spatial similarity into active learning, an adaptive superpixel (SP) segmentation algorithm, which can save SPs in uninformative regions and preserve edge details in complex regions, is employed to generate better local spatial constraints for AL. Quantitative and qualitative results indicate that the MAT-ASSAL outperforms seven state-of-the-art methods on three HSI datasets.

Convolutional neural networks are widely used in the field of hyperspectral image classification because of their excellent nonlinear feature extraction ability. However, as the sampling position of the regular convolution kernel is unchangeable, the regular convolution cannot distinctively extract the spatial and spectral information around the central pixel, which makes the classification results at the boundaries of ground objects over-smoothed and the classification performance degraded. Thus, we propose a novel superpixel guided deformable convolution network (SGDCN) for hyperspectral image classification. Firstly, the superpixel region fusion filter (SRF-Filter) is designed to fuse the initial superpixel region segmented by the simple linear iterative clustering (SLIC), making the fused superpixel region have a high homogeneity and also contain spatial features of diverse scales. Then, the superpixel guided deformable convolution (SGD-Conv) is proposed to make the shape of deformable convolution consistent with the real shape of land covers, and the SGD-Conv can extract pure neighborhood spatial-spectral features. Finally, a superpixel joint bilateral filter (SPJBF) is designed to solve the pixel-level and region-level misclassification problem, which can effectively utilize the superpixel region’s homogeneity and improve the classification accuracy. Experiments on three HSI datasets indicate that the SGDCN can obtain better classification performance when compared with other twelve state-of-the-art methods.

Cross-scene hyperspectral image classification (HSIC) poses a significant challenge in recognizing hyperspectral images (HSIs) from different domains. The current mainstream approaches based on domain adaptation (DA) methods need to access target data when aligning distributions between domains, limiting the applicability of the model. In contrast, recent domain generalization (DG) methods aim to directly generalize to unseen domains, eliminating the requirements for target data during training. Nonetheless, most DG-based methods overly focus on randomizing sample styles, leading to semantically compromised samples. In addition, broadening the source distribution without ensuring reasonable support may result in undesired extended distributions. To address these issues, we propose a novel DG network with frequency disentanglement and data geometry (FDGNet) for cross-scene HSIC. Specifically, we first develop a spectral–spatial encoder based on frequency disentanglement (FDSS encoder), which facilitates synthesized domains to preserve their semantic consistency while simulating interdomain gaps with the source domain. Second, to avoid the generation of unrealistic samples, we incorporate data geometry into adversarial training. This helps diversify new domains while keeping the data geometry of extended domains in an explainable support. To improve the learning of domain-invariant representation, we propose an intermediate domain sampling strategy based on the class-wise perceptual manifold. This strategy synthesizes reliable intermediate domains by sampling from class-wise manifold flows estimated over the source and extended domains. Extensive experiments and analysis on three public HSI datasets yield the superiority of our proposed FDGNet. The codes will be available from the website: https://github.com/Qba-heu/FDGNet.

The Convolutional Neural Network (CNN) has been used in many fields and has achieved remarkable results, such as image classification, face detection, and speech recognition. Compared to GPU (graphics processing unit) and ASIC, a FPGA (field programmable gate array)-based CNN accelerator has great advantages due to its low power consumption and reconfigurable property. However, FPGA’s extremely limited resources and CNN’s huge amount of parameters and computational complexity pose great challenges to the design. Based on the ZYNQ heterogeneous platform and the coordination of resource and bandwidth issues with the roofline model, the CNN accelerator we designed can accelerate both standard convolution and depthwise separable convolution with a high hardware resource rate. The accelerator can handle network layers of different scales through parameter configuration and maximizes bandwidth and achieves full pipelined by using a data stream interface and ping-pong on-chip cache. The experimental results show that the accelerator designed in this paper can achieve 17.11GOPS for 32bit floating point when it can also accelerate depthwise separable convolution, which has obvious advantages compared with other designs.

Existing hyperspectral cross-domain few-shot learning (FSL) methods focus mainly on elaborating on training strategies or domain alignment algorithms, while paying less attention to the biased metaknowledge introduced by a large amount of source data and the implicit encouragement of learning target domain-specific attributes. In this article, from the perspective of disentangled representation learning, a novel cross-domain FSL method based on feature disentanglement (FDFSL) is proposed for hyperspectral image classification (HSIC). Specifically, to suppress the representation biased toward the source data and enable the model to implicitly focus on the inherent knowledge of the target domain, an orthogonal low-rank feature disentanglement method is employed to acquire desired features of source and target pipelines. Furthermore, to preserve more shared and discriminative information from the heterogeneous data space (i.e., the spectral dimensions of the source and target scenes are typically different), a multiorder spectral interaction block based on central position encoding (MICD) is proposed to fully integrate the respective features into the spectral domain, which allows the model to emphasize informative spectral dimensions in a data-driven manner. Finally, to diversify the feature representation space while preventing the model overfitting to domain alignment task, a self-distillation scheme is developed to facilitate the acquisition of task-relevant feature components. Extensive experiments and analysis on three public HSI datasets suggest the superiority of the proposed method. The code will be available on the website at https://github.com/Qba-heu/FDFSL .