Solar-induced chlorophyll fluorescence (SIF) is a direct proxy for photosynthetic activity, yet existing satellite SIF products are constrained by coarse spatial resolution, limiting their application in ecological and agricultural studies. In this work, we propose a One-Step Learned Spatial Redistribution Convolutional Neural Network (OSRNet) that downscales 0.05° TROPOMI SIF to 0.005° by learning spatially adaptive redistribution fields from high-resolution drivers, which allocate coarse-resolution satellite SIF into fine-resolution grids. Based on this framework, we generate RSIF, a global 16-day 0.005° SIF dataset for 2018–2020. Comprehensive evaluation against both satellite and tower-based SIF shows that RSIF maintains strong consistency with TROPOMI observations (R² = 0.976, RMSE = 0.036) while recovering fine-scale spatial details. OSRNet substantially outperforms established direct prediction methods such as RF and SIFNet, and, compared with post hoc corrected RF approach from prior studies, achieves the highest R² across all tower sites and generally the lowest RMSE, enabling more accurate representation of seasonal dynamics with improved spatial fidelity.

Hyperspectral imaging enables advanced change detection but struggles with extensive redundant data across spatial and spectral dimensions. This bloats model size and computational loads. To address this problem, we propose a new gated spectral–spatial–temporal attention network with spectral similarity filtering (HyGSTAN) with a lightweight yet accurate architectural design. Specifically, our HyGSTAN introduces three innovative modules: 1) spectral similarity filtering to reduce spectral redundancy via cosine similarity; 2) gated spectral-spatial attention to capture intra-image spatial features using single-head weak self-attention and gated mechanisms; and 3) gated spectral–spatial–temporal attention to extract inter-image temporal changes. Experiments on three benchmark datasets demonstrate HyGSTAN’s ability to balance accuracy, model complexity, and computational efficiency. The proposed attention mechanisms extract more discriminative information without sacrificing performance. The source code of this work will be released at https://github.com/Welcome-to-LISA/HyGSTAN.

Thermal infrared remote sensing has been widely used for sea surface temperature (SST) monitoring. However, traditional SST remote sensing retrieval models usually have uncertainties when applied to offshore waters with variable environmental conditions due to simple functional forms and empirical fitting of model parameters. Machine learning (ML) can theoretically avoid these problems, but its applicability and driving features have not been thoroughly investigated. Here, we proposed a high-precision ML-based SST retrieval framework for offshore waters in the eastern Liaodong Peninsula of China by optimizing the selection of input features. The results showed that random forest model achieved better accuracy than deep neural network and the improved split-window algorithm, exhibited credible spatial patterns of SST maps across four seasons, and was portable for the independent samples in 2021. This study offers references in selection of features and models for SST retrieval, and benefits the accuracy of offshore SST retrieval.

Due to the powerful feature information mining ability of deep learning, models such as Convolutional Neural Network (CNN) and Transformer have gained a certain progress in hyperspectral image classification (HSIC). Characteristically, the CNN is good at extracting local information, but it has the limitation of insufficient receptive field. While the Transformer has the advantage of global representation, it ignores local details to some extent. Therefore, this letter proposes an interactive Transformer and CNN with multilevel feature fusion network (ITCNet) for HSIC. Specifically, in the image-based framework, features with different perceptual fields and depths are extracted interactively by a multi-layer Transformer and CNN, then fused through a multilevel feature fusion module for class prediction. Experimental results on two real datasets verifies its efficiency, with improvements over other related methods.

Detecting and counting solar panels from high-resolution aerial images timely and accurately is essential for monitoring and management of industrial solar photovoltaic (PV) systems. Due to the influence of weather and light, the detection results of traditional methods are usually unsatisfactory. For the purpose of improving detection accuracy, we propose a method that combined residual network and channel attention module to improve the Faster RCNN framework. First, the balance between the training data size and model complexity is investigated, and the residual network is utilized to deepen the feature extractor within the effective range. Then, the channel attention modules are introduced into the network to further enhance the feature representation. Experimental results, conducted on high-resolution aerial image over Guilin, China, prove that the proposed method can detect solar panels with better accuracy than other related methods.

Sparse representation (SR)-based models have shown to be a powerful category of frameworks for hyperspectral image classification (HSIC). However, current residual-driven methods mainly focus on the sparsity of the coefficient, which is generally used in conjunction with the dictionary. In fact, the discriminant information hidden behind the value of sparse coefficient is not fully exploited. In this letter, we analyze the SR-based framework from the perspective of sparse coefficient, develop the participation degree (PD)-driven decision mechanism, and establish a concise model called constraint representation (CR). Based on CR, an improved version called adjacent CR (ACR) is further proposed, with consideration of spatial coherence via adjacent constraint. Experimental results using two real hyperspectral datasets verify the improvements of the proposed methods over the other related models and their spatial variants.

Sparse representation (SR)-based models have been widely applied for hyperspectral image classification. However, the original residual-driven frameworks ignore the property of sparse coefficient to some extent, and their spatial variants suffer obstacles of optimization due to the strong constraint. In this paper, based on previous works on sparse coefficient and its spatial expansion, we put forward a novel classifier, called superpixel-level constraint representation (SPCR). In particular, constraint representation (CR) is first applied to interpret the process of SR from perspective of participation degree (PD). Then, a relaxed and adaptive spatial constraint via superpixel segmentation is imposed to transform the individual PD to local relative activity degree (RAD). The final classification is determined based on a concise RAD-driven mechanism. Experimental results on real data set demonstrate the efficiency of the proposed method.