Convolutional neural networks (CNNs) and transformer have achieved excellent classification performances in HSI and LiDAR land cover classification. However, for complex land covers, effectively characterizing the contextual information and spectral-spatial interaction features of HSI and LiDAR is crucial for improving classification accuracy. Motivated by this, this paper is dedicated to selective convolutional kernel mechanisms and spectral-spatial interactive transformer feature learning style, proposing a selective spectral-spatial aggregation transformer network, named S2ATNet. A convolution feature selected module (CFSM), which can dynamically capture its contextual features of various land covers, is firstly utilized into both of HSI and LiDAR branches. Afterwards, a cascaded spatial-spectral learning and interactive fusion block (CSLIF) is designed for acquiring the non-local spatial-spectral characteristics in an interactive feature learning style. The learned features are fed into max-average classification head (MACH) to obtain the final classification results. The effectiveness of the proposed S2ATNet is validated on two publicly available datasets. Codes are available at https://github.com/RSIP-NJUPT/S2ATNet.git.

Benefiting from the inherent capability of graph convolutional neural networks (GCNs) to flexibly convolve over regions with arbitrary shapes, they could model the spatial topology of SAR land covers. However, existing GCNs-based classification methods often work on a static graph and perform graph average pooling which have the fixed weights as a node pooling style, thus lacking the dynamic graph pooling representation of SAR patches under complex spatial structure. Moreover, distinguishable feature learning under different GCN blocks is generally ignored, resulting in GCNs-based methods struggling to exploit intrinsic topological structure of SAR land covers across all scales. To alleviate these issues, we propose a Dynamic Pyramidal Graph U-Net (DPGUNet), including residual graph convolution modulate fusion layer (RGCMF), attentional node feature aggregation pooling layer (ANFAP), and graph pyramidal feature learning module (GPFL), for leveraging spatial topological structures and hierarchical semantic features of SAR patches effectively. To be specific, RGCMF performs SAR global and local feature fusion between the pre-activation graph convolution and its residual branch via graph attention feature fusion module (GAFF); ANFAP is employed to transfer node information between dynamic graphs in pooling stage; Furthermore, we utilize a GPFL, which is defined by cross-scale merging association matrices, to acquire multi-scale spatial topological and contextual semantic features of SAR patches. Experimental results on three real SAR land cover classification datasets demonstrate the superiority of DPGUNet compared to several other related methods. Code is available at https://github.com/RSIP-NJUPT/DPGUNet .

The transformer-based methods have demonstrated remarkable advancements in synthetic aperture radar (SAR) classification. Nevertheless, many of these methods ignore global statistical information and semantic feature interaction for effectively characterizing different SAR land covers under complex structure. Leveraging second-order statistics presents an efficacious approach to characterize statistical features of SAR patches well. Motivated by this, we integrate pyramid pooling and global covariance pooling techniques into each of the multi-head self-attention (MHSA) blocks, thereby facilitating the extraction of powerful contextual features and global statistical nature of SAR patches, namely multi-order pooling transformer module (MPTM). Simultaneously, a semantic feature aggregation module (SFAM) is utilized for capturing local deep features and modeling the interaction of feature information across various feature levels. Both these modules are embedded into a U-shaped architecture, which we refer to as multi-order pooling transformer-based semantic feature aggregation network (MPT-SFANet). Through extensive experimental results on TerraSAR, Sentinel-1B, and GF-3 SAR image classification datasets indicate that MPT-SFANet exceeds several relevant methods.

The description of context information affected by speckle and class imbalance under labeled data makes the pixelwise classification for high-resolution (HR) synthetic aperture radar (SAR) image a challenging task. To address these issues, we propose a global-context pyramidal and class-balanced network (GPCNet) for HR SAR image classification. The proposed structure follows an encoder–decoder architecture. In the encoder module, the multiscale convolutional and global-local cross-channel attention (GCA) blocks are employed to capture the global-context and distinguishable deep feature statistics, while reducing the impacts of the random fluctuation in the homogeneous region. The channel information of different scale convolutional layers is efficiently learned by local cross-channel interaction in the GCA block. Besides, a sampled class-balanced loss, associating with the effective number, is utilized for alleviating the class imbalance of HR SAR images. The experiments carried out on a TerraSAR-X image classification dataset demonstrate GPCNet is able to yield superior performance when compared with other related networks.

Land Surface Temperature (LST) is a critical parameter for environmental studies, but directly obtaining high spatial resolution LST data remains challenging due to the spatio-temporal trade-off in satellite remote sensing. Guided LST downscaling has emerged as an alternative solution to overcome these limitations, but current methods often neglect spatial non-stationarity, and there is a lack of an open-source ecosystem for deep learning methods. In this paper, we propose the Modality-Conditional Large Selective Kernel (MoCoLSK) Network, a novel architecture that dynamically fuses multi-modal data through modality-conditioned projections. MoCoLSK achieves a confluence of dynamic receptive field adjustment and multi-modal feature fusion, leading to enhanced LST prediction accuracy. Furthermore, we establish the GrokLST project, a comprehensive open-source ecosystem featuring the GrokLST dataset, a high-resolution benchmark, and the GrokLST toolkit, an open-source PyTorch-based toolkit encapsulating MoCoLSK alongside 40+ state-of-the-art approaches. Extensive experimental results validate MoCoLSK’s effectiveness in capturing complex dependencies and subtle variations within multispectral data, outperforming existing methods in LST downscaling. Our code, dataset, and toolkit are available at https://github.com/GrokCV/GrokLST.

Due to the effective embedding of temporal information, time-series SAR images can acquire more abundant land covers information, thereby effectively enhancing the accuracy of land cover mapping. However, the significant introduction of temporal features greatly increases the redundancy of deep features. Simultaneously, capturing the global-local features of time-series SAR images effectively while considering temporal features has a significant influence on time-series SAR land cover mapping. Motivated by these, a sparse 3D Transformer network based on joint learning of temporal-spatial features, named Sparse Local-Global Temporal Transformer Network (SL-GT2Net), is proposed for SAR land cover mapping, which has the ability to learn sparse global-local temporal-spatial features of SAR land covers. To emphasize crucial spatiotemporal features and suppress noise features, we propose a Spatial-Temporal Channel Refinement Block (STCRB) to mitigate feature redundancy. Additionally, we develop novel and effective Sparse Multi-scale Local Window Multi-Head Self-Attention (SMLW-MHSA) and Sparse Multi-scale Global Window Multi-Head Self-Attention (SMGW-MHSA) blocks. Both incorporate parallel multi-scale window partitioning and Top-K sparsification strategies to robustly model discriminative spatiotemporal features of land covers with vast scale variations, while maintaining linear complexity with respect to the number of tokens. The experimental results on three challenge time-series SAR datasets demonstrate that our SL-GT2Net exhibits outstanding competitiveness compared to other related networks. The code is available at https://github.com/RSIP-NJUPT/SL-GT2Net.