Cellular traffic prediction is crucial for optimizing network resources and enhancing service quality. Despite progress in existing traffic prediction methods, challenges remain in capturing periodic features, spatial heterogeneity, and abnormal signals. To address these challenges, we propose a Station-aware Graph Attention Sequence Network (SGA-Seq). The core idea is to achieve accurate cellular traffic prediction by adaptively modeling station-specific spatiotemporal patterns and effectively handling complex traffic dynamics. First, we introduce a learnable temporal embedding mechanism to capture temporal features across multiple scales. Second, we design a station-aware graph attention network to model complex spatial relationships across stations. Additionally, by progressively separating regular and abnormal signals layer by layer, we enhance the model’s robustness. Experimental results demonstrate that SGA-Seq outperforms existing methods on five diverse mobile network datasets spanning different scales, including cellular traffic, mobility flow, and communication datasets. Notably, on the V-GCT dataset, our method achieves an 8.04% improvement in Root Mean Squared Error compared to the Spatiotemporal-aware Trend-Seasonality Decomposition Network. The code of SGA-Seq is available at https://github.com/OvOYu/SGA-Seq.

Sea ice change detection is vital for understanding climate dynamics and ensuring maritime safety. Existing deep learning methods often struggle with the significant impact of color variations in satellite imagery, which can lead to inaccurate detection results. Moreover, the scarcity of labeled sea ice change data limits the ability of models to generalize across diverse scenarios. To address these challenges, we propose SICNet, a sea ice change detection model with enhanced color robustness and data efficiency. A Wavelet-guided Color-robust Fusion module is introduced to reduce low-frequency color discrepancies while preserving high-frequency edge details. Additionally, a novel Change-Sensitive CutMix strategy is employed to augment training samples by focusing on regions with moderate change, effectively increasing data diversity. Experiments conducted on our constructed sea ice change dataset demonstrate that SICNet achieves superior performance and robustness under varying environmental and lighting conditions. The source code of SICNet is available at https://github.com/viking-hong/SICNet.git.

Arbitrary-oriented object detection (AOOD) has been widely applied to locate and classify objects with diverse orientations in remote sensing images. However, the inconsistent features for the localization and classification tasks in AOOD models may lead to ambiguity and low-quality object predictions, which constrains the detection performance. In this article, an AOOD method called task-wise sampling convolutions (TS-Conv) is proposed. TS-Conv adaptively samples task-wise features from respective sensitive regions and maps these features together in alignment to guide a dynamic label assignment for better predictions. Specifically, sampling positions of the localization convolution in TS-Conv are supervised by the oriented bounding box (OBB) prediction associated with spatial coordinates, while sampling positions and convolutional kernel of the classification convolution are designed to be adaptively adjusted according to different orientations for improving the orientation robustness of features. Furthermore, a dynamic task-consistent-aware label assignment (DTLA) strategy is developed to select optimal candidate positions and assign labels dynamically according to ranked task-aware scores obtained from TS-Conv. Extensive experiments on several public datasets covering multiple scenes, multimodal images, and multiple categories of objects demonstrate the effectiveness, scalability, and superior performance of the proposed TS-Conv.

This paper presents a new dual domain network with image prior based on Convolutional Neural Network (CNN) and Transformer simultaneously for CT Metal Artifact Reduction (MAR). Challenges in MAR derive from the following aspects: firstly, the different morphologies of metal artifacts complexify resolving the issue just in a single domain; secondly, albeit many methods excel in quantitative metrics, yet the restored anatomical structures are over-smooth blurring reconstructed CT images; thirdly, MAR demands better performance as a clinical application, but the approaches relying on CNN or Transformer struggle due to CNN’s restricted spatial scope and Transformer’s ignorance to the local details, respectively, that is, CNN focuses on the local information while Transformer emphasizes the global information with higher computational complexity. To address these problems, we put forward MAReraser, a novel dual domain network, to deal with metal artifacts. MAReraser removes metal artifacts in both the projection and image domains, effectively reducing heteromorphic metal artifacts. Moreover, MAReraser introduces the image prior generated by an image prior subnet to refine the quality of reconstructed CT images. The prior subnet is pretrained in an expanded dataset which incorporates CT images corrected by diverse traditional MAR methods, providing extra potential prior knowledge from different perspectives. Further, the network backbone of MAReraser integrates CNN and Transformer, enabling complementary local and global feature extraction and balancing computational complexity. Extensive experiment results demonstrate that our method outperforms several other approaches whether in quantitative metrics or in qualitative visualization results.

The recognition of sea ice is of great significance for reflecting climate change and ensuring the safety of ship navigation. Recently, many deep learning based methods have been proposed and applied to segment and recognize sea ice regions. However, the diverse scales of sea ice areas, the zigzag and fine edge contours, and the difficulty in distinguishing different types of sea ice pose challenges to existing sea ice recognition models. In this paper, a Global-Local Detail Guided Transformer (GDGT) method is proposed for sea ice recognition in optical remote sensing images. In GDGT, a global-local feature fusiont mechanism is designed to fuse global structural correlation features and local spatial detail features. Furthermore, a detail-guided decoder is developed to retain more high-resolution detail information during feature reconstruction for improving the performance of sea ice recognition. Experiments on the produced sea ice dataset demonstrated the effectiveness and advancement of GDGT.

The recognition of sea ice is of great significance for reflecting climate change and ensuring the safety of ship navigation. Recently, many deep-learning-based methods have been proposed and applied to segment and recognize sea ice regions. However, there are huge differences in sea ice size and irregular edge profiles, which bring challenges to the existing sea ice recognition. In this article, a global-local Transformer network, called SeaIceNet, is proposed for sea ice recognition in optical remote sensing images. In SeaIceNet, a dual global-attention head (DGAH) is proposed to capture global information. On this basis, a global-local feature fusion (GLFF) mechanism is designed to fuse global structural correlation features and local spatial detail features. Furthermore, a detail-guided decoder is developed to retain more high-resolution detail information during feature reconstruction for improving the performance of sea ice recognition. Extensive experiments on several sea ice datasets demonstrated that the proposed SeaIceNet has better performance than the existing methods in multiple evaluation indicators. Moreover, it excels in addressing challenges associated with sea ice recognition in optical remote sensing images, including the difficulty in accurately identifying irregular frozen ponds in complex environments, the broken and unclear boundaries between sea and thin ice that hinder precise segmentation, and the loss of high-resolution spatial details during model learning that complicates refinement.

Subsurface density (SD) is a crucial dynamic environment parameter reflecting a 3-D ocean process and stratification, with significant implications for the physical, chemical, and biological processes of the ocean environment. Thus, accurate SD retrieval is essential for studying dynamic processes in the ocean interior. However, complete spatiotemporally accurate SD retrieval remains a challenge in terms of the equation of state and physical methods. This study proposes a novel multiscale mixed residual transformer (MMRT) neural network method to compensate for the inadequacy of the existing methods in dealing with spatiotemporal nonlinear processes and dependence. Considering the spatial correlation and temporal dependence of dynamic processes within the ocean, the MMRT addresses temporal dependence by fully using the transformer’s processing of time-series data and spatial correlation by compensating for deficiencies in spatial feature information through multiscale mixed residuals. The MMRT model was compared with the existing random forest (RF) and recurrent neural network (RNN) methods. The MMRT model achieves the best accuracy with an average determination coefficient (${R}^{2}$) of 0.988 and an average root mean square error (RMSE) of 0.050 kg/m3 for all layers. The MMRT model not only outperforms the RF and RNN methods regarding reliability and generalization ability when estimating global ocean SD from remote sensing data but also has a more interpretable encoding process. The MMRT model offers a new method for directly estimating SD using multisource satellite observations, providing significant technical support for future remote sensing super-resolution and prediction of subsurface parameters.

3-D ocean temperature and salinity data are the basis for studying ocean dynamic processes and warming. Satellite remote sensing observations on the ocean surface are abundant and full-coverage, while in situ observations in the ocean interior are very sparse and unevenly distributed. Currently, the remote sensing inversion models of temperature and salinity in the ocean interior are unable to learn both global and local detail information, and modeling layer-by-layer blocks the connection between vertical depth levels, resulting in poor accuracy. In this study, we proposed a novel clustering-guided and knowledge-distillation network (CGKDN) model based on the ocean knowledge-driven model. The model introduced K-means clustering for the partitions of ocean processes, knowledge distillation (KD) fusing global and local detail information, and adaptive depth gradient loss linking the vertical depth dimension, which enhanced the interpretability and accuracy of the model. Comparison of the reconstructions with the existing major publicly available datasets through the validation of 10% EN4 in situ profile observations from 2001 to 2020 reveals that the reconstructions are more accurate. Concretely, the average root mean square error (RMSE) (°C) across time-series and vertical levels of CGKDN/Institute of Atmospheric Physics (IAP)/OCEAN5 ocean analysis-reanalysis (ORAS5)/deep ocean remote sensing (DORS) ocean subsurface temperature (OST) is 0.590/0.598/0.690/0.723, and the average RMSE (PSU) of CGKDN/IAP/ORAS5 ocean subsurface salinity (OSS) is 0.101/0.103/0.106, respectively. Furthermore, the downscaled quarter-degree reconstructions present more mesoscale detail signals, consistent with the ARMOR3D data. This study not only improves the estimation accuracy of subsurface temperature and salinity but also serves the study of ocean interior dynamic processes and variabilities and provides valuable references for reconstructing other ocean subsurface physical variables.

In large-scale disaster events, the planning of optimal rescue routes depends on the object detection ability at the disaster scene, with one of the main challenges being the presence of dense and occluded objects. Existing methods, which are typically based on the RGB modality, struggle to distinguish targets with similar colors and textures in crowded environments and are unable to identify obscured objects. To this end, we first construct two multimodal dense and occlusion vehicle detection datasets for large-scale events, utilizing RGB and height map modalities. Based on these datasets, we propose a multimodal collaboration network for dense and occluded vehicle detection, MuDet for short. MuDet hierarchically enhances the completeness of discriminable information within and across modalities and differentiates between simple and complex samples. MuDet includes three main modules: Unimodal Feature Hierarchical Enhancement (Uni-Enh), Multimodal Cross Learning (Mul-Lea), and Hard-easy Discriminative (He-Dis) Pattern. Uni-Enh and Mul-Lea enhance the features within each modality and facilitate the cross-integration of features from two heterogeneous modalities. He-Dis effectively separates densely occluded vehicle targets with significant intra-class differences and minimal inter-class differences by defining and thresholding confidence values, thereby suppressing the complex background. Experimental results on two re-labeled multimodal benchmark datasets, the 4K-SAI-LCS dataset, and the ISPRS Potsdam dataset, demonstrate the robustness and generalization of the MuDet.

Recently, many arbitrary-oriented object detection (AOOD) methods have been proposed and applied to remote sensing and other fields. For aerial platforms, lightweight structure and multimodal adaptations of convolutional neural network (CNN) models are urgently needed. Due to the limited model size, the performance of existing lightweight AOOD methods is low, especially in multimodal tasks. In this paper, a multimodal knowledge distillation (MKD) method is proposed for AOOD in aerial images. In MKD, a multimodal dynamic label assignment strategy is designed to select the optimal positive samples dynamically to adapt to different modalities and environments. Different multimodal localization and feature distillation modules are designed to make multimodal knowledge to be complementary and effectively learned by the lightweight model. Experiments on the public dataset demonstrated the effectiveness and advancement of MKD.

Recently, many lightweight neural networks have been deployed on airborne or satellite remote sensing platforms for real-time object detection. To bridge the performance gap between lightweight models and complex models, many knowledge distillation (KD) methods are investigated. However, existing KD methods ignore to transfer effective directional knowledge. Meanwhile, knowledge of different subtasks interfere with each other. To this end, a directional alignment instance knowledge distillation (DAIK) method for improving the performance of the lightweight object detection model is proposed. Specifically, an angle distillation (AD) module is developed to combine the circular smooth label and teacher logits to transfer effective directional knowledge. Angular-distance Aspect-ratio Look-up-table (AAL) is incorporated into label assignment and re-weighting loss to enhance the prediction sensitivity of direction and shape in a discrete manner. Sample alignment distillation (SAD) reduces the spatial misalignment by mimicking the teacher model’s distribution of anchor points. Extensive experiments are performed on several public remote sensing object detection datasets, which demonstrates the effectiveness of the proposed DAIK.

Recently, some lightweight convolutional neural network (CNN) models have been proposed for airborne or spaceborne remote sensing object detection (RSOD) tasks. However, these lightweight detectors suffer from performance degradation due to the compromise of limited computing resources on embedded devices. In order to narrow this performance gap, a direction-adaptive knowledge extraction and distillation (DKED) method is proposed. Specifically, a dynamic directional convolution (DDC) is developed to extract the typical arbitrary-oriented features, and a direction-adaptive knowledge distillation (DKD) strategy is designed for guiding the lightweight model to learn the intrinsic knowledge of the RSOD task from the high-performance model. Experiments on public datasets demonstrate that the proposed method can effectively improve the performance of the lightweight RSOD model without additional inference costs.

Infrared small target detection (ISTD) has attracted widespread attention and been applied in various fields. Due to the small size of infrared targets and the noise interference from complex backgrounds, the performance of ISTD using convolutional neural networks (CNNs) is restricted. Moreover, the constriant that long-distance dependent features can not be encoded by the vanilla CNNs also impairs the robustness of capturing targets' shapes and locations in complex scenarios. To this end, a multi-patch attention network (MPANet) based on the axial-attention encoder and the multi-scale patch branch (MSPB) structure is proposed. Specially, an axial-attention-improved encoder architecture is designed to highlight the effective features of small targets and suppress background noises. Furthermore, the developed MSPB structure fuses the coarse-grained and fine-grained features from different semantic scales. Extensive experiments on the SIRST dataset show the superiority performance and effectiveness of the proposed MPANet compared to the state-of-the-art methods.

With the gradual opening of airspace, tracking of noncooperative low-altitude slow-speed small size (LSS) targets is important for the maintenance of security. It is still a challenging problem, especially for complex scenarios and real-time constraints. In this letter, an efficient tracking by relocalization (TRL) framework is proposed for small flying object tracking, aiming to alleviate the issue of losing moving targets in a complex background. Our designed relocalization module consists of a feature-aggregated module and a global search module. On the one hand, a feature-aggregated module is integrated into the designed framework to increase the ability to locate small targets. On the other hand, a global search module is developed to update the tracking performance, which attempts to address missed targets in long-term small object tracking tasks. What needs to be declared is that the basic tracking module cooperates with the relocalization module we designed to achieve the tracking of small targets. Performance evaluation of two small-flying target data sets and comparison with several state-of-the-art approaches demonstrate the effectiveness of the proposed framework.

Infrared objects acquired from a long-distance have small sizes and are easily submerged by a complex and variable background. The existing deep network detection framework suffers greatly from the feature spatial resolution loss caused by the networks’ depth and multiple downsampling operations, which is extremely detrimental for small object detection. So, a crucial and urgent goal is, how to trade-off network depth and feature spatial resolution, while learning feature context representation and interaction to distinguish from the background. To this end, we propose a deep interactive U-Net architecture (short for DI-U-Net) with high feature learning and feature interaction ability. First, feature learning is first achieved through a multi-level and high-resolution network structure. This structure ensures feature resolution as the network depth increase, and also focus on the object’s global context information. Then, the feature interactive is further achieved by the dense feature encoder (DFI) module to learn object local context information. The proposed method yields strong object context representation and well discriminability, as well as a good fit for infrared small object detection. Extensive experiments are conducted on the SISRT dataset and Synthetic dataset, demonstrating the superiority and effectiveness of the proposed deeper U-Net compared to previous state-of-the-art detection methods.

Object detection (OD) is an important task of computer vision and has been widely used in many fields, including remote sensing (RS). However, the complex scenes, large-scale variation, and dense instances of RS bring huge challenges to OD. To meet these challenges, a novel Nonlocal-aware Pyramid and Multiscale Multitask Refinement Detector (NPMMR-Det) is proposed. Specifically, nonlocal-aware pyramid attention (NP-Attention) is designed for guiding a neural network model to focus more on efficient features and suppress background noise. Then a multiscale refinement feature pyramid network (MSR-FPN) is proposed to fuse the multiscale context features extracted by the NP-Attention guided neural network and adjust the optimal receptive field. In order to use these features more effectively, a multitask refinement head called MTR-Head, with offset sharing and a modulation mechanism, is developed to refine the feature misalignment between the localization task and the classification task. Extensive experiments performed on two public RS data sets demonstrate that the proposed NPMMR-Det achieves competitive performance compared with state-of-the-art methods.

A few lightweight convolutional neural network (CNN) models have been recently designed for remote sensing object detection (RSOD). However, most of them simply replace vanilla convolutions with stacked separable convolutions (SConvs), which may not be efficient due to a lot of precision losses and may not be able to detect oriented bounding boxes (OBBs). Also, the existing OBB detection methods are difficult to constrain the shape of objects predicted by CNNs accurately. In this article, we propose an effective lightweight oriented object detector (LO-Det). Specifically, a channel separation-aggregation (CSA) structure is designed to simplify the complexity of SConvs, and a dynamic receptive field (DRF) mechanism is developed to maintain high accuracy by customizing the convolution kernel and its perception range dynamically when reducing the network complexity. The CSA-DRF component optimizes efficiency while maintaining high accuracy. Then, a diagonal support constraint head (DSC-Head) component is designed to detect OBBs and constrain their shapes more accurately and stably. Extensive experiments on public data sets demonstrate that the proposed LO-Det can run very fast even on embedded devices with the competitive accuracy of detecting oriented objects.

Arbitrary-oriented object detection (AOOD) plays a significant role in image understanding in remote-sensing scenarios. The existing AOOD methods face the challenges of ambiguity and high costs in angle representation. To this end, a multigrained angle representation (MGAR) method, consisting of coarse-grained angle classification (CAC) and fine-grained angle regression (FAR), is proposed. Specifically, the designed CAC avoids the ambiguity of angle prediction by discrete angular encoding (DAE) and reduces complexity by coarsening the granularity of DAE. Based on CAC, FAR is developed to refine the angle prediction with much lower costs than narrowing the granularity of DAE. Furthermore, an Intersection over Union (IoU)-aware FAR-Loss (IFL) is designed to improve the accuracy of angle prediction using an adaptive reweighting mechanism guided by IoU. Extensive experiments are performed on several public remote-sensing datasets, which demonstrate the effectiveness of the proposed MGAR. Moreover, experiments on embedded devices demonstrate that the proposed MGAR is also friendly for lightweight deployments.

Recently, many arbitrary-oriented object detection (AOOD) methods have been proposed and attracted widespread attention in many fields. However, most of them are based on anchor-boxes or standard Gaussian heatmaps. Such label assignment strategy may not only fail to reflect the shape and direction characteristics of arbitrary-oriented objects, but also have high parameter-tuning efforts. In this paper, a novel AOOD method called General Gaussian Heatmap Label Assignment (GGHL) is proposed. Specifically, an anchor-free object-adaptation label assignment (OLA) strategy is presented to define the positive candidates based on two-dimensional (2D) oriented Gaussian heatmaps, which reflect the shape and direction features of arbitrary-oriented objects. Based on OLA, an oriented-bounding-box (OBB) representation component (ORC) is developed to indicate OBBs and adjust the Gaussian center prior weights to fit the characteristics of different objects adaptively through neural network learning. Moreover, a joint-optimization loss (JOL) with area normalization and dynamic confidence weighting is designed to refine the misalign optimal results of different subtasks. Extensive experiments on public datasets demonstrate that the proposed GGHL improves the AOOD performance with low parameter-tuning and time costs. Furthermore, it is generally applicable to most AOOD methods to improve their performance including lightweight models on embedded platforms.