Deep learning has recently achieved remarkable progress in hyperspectral image (HSI) classification. Among these advancements, the Transformer-based models have gained considerable attention due to their ability to establish long-range dependencies. However, the quadratic computational complexity of the self-attention mechanism limits its application in hyperspectral image classification (HSIC). Recently, the Mamba architecture has shown outstanding performance in 1D sequence modeling tasks owing to its lightweight linear sequence operations and efficient parallel scanning capabilities. Nevertheless, its application in HSI classification still faces challenges. Most existing Mamba-based approaches adopt various selective scanning strategies for HSI serialization, ensuring the adjacency of scanning sequences to enhance spatial continuity. However, these methods lead to substantially increased computational overhead. To overcome these challenges, this study proposes the Hyperspectral Spatial Mamba (HyperSMamba) model for HSIC, aiming to reduce computational complexity while improving classification performance. The suggested framework consists of the following key components: (1) a Multi-Scale Spatial Mamba (MS-Mamba) encoder, which refines the state-space model (SSM) computation by incorporating a Multi-Scale State Fusion Module (MSFM) after the state transition equations of original SSMs. This module aggregates adjacent state representations to reinforce spatial dependencies among local features; (2) our proposed Adaptive Fusion Attention Module (AFAttention) to dynamically fuse bidirectional Mamba outputs for optimizing feature representation. Experiments were performed on three HSI datasets, and the findings demonstrate that HyperSMamba attains overall accuracy of 94.86%, 97.72%, and 97.38% on the Indian Pines, Pavia University, and Salinas datasets, while maintaining low computational complexity. These results confirm the model’s effectiveness and potential for practical application in HSIC tasks.

The U-net network, with its simple and powerful encoder–decoder structure, dominates the field of medical image segmentation. However, convolution operations are limited by receptive fields. They do not have the ability to model long-range dependencies, but Transformer has the capability of long-term modeling thanks to its core self-attention mechanism, which has been widely applied in the field of medical image segmentation. However, both CNNs and Transformer can only perform correlation calculations for a single sample, ignoring the correlation between different samples. To address these problems, we propose a new Transformer, which we call the Dual-Attention Transformer (DAT). This module captures correlations within a single sample while also learning correlations between different samples. The current U-net and some of its variant models have the problem of inadequate feature fusion, so we also improve the skip connection to strengthen the association between feature maps at different scales, reduce the semantic gap between the encoder and decoder, and further improve the segmentation performance. We refer to this structure as DATUnet. We conducted extensive experiments on the Synapse and ACDC datasets to validate the superior performance of our network, and we achieved an average DSC (%) of 83.6 and 90.9 and an average HD95 of 13.99 and 1.466 for the Synapse and ACDC datasets, respectively.

In recent years, unmanned aerial vehicles (UAVs) have gradually become an important tool for forest fire monitoring due to their flexibility and wide-area observation capability. However, existing detection algorithms still struggle to achieve a balance between computational complexity and detection performance in diverse background conditions, and data for such scenarios remain limited. Therefore, this paper proposes a lightweight forest fire smoke detection model based on YOLOv10n. Specifically, we introduce the RepViTBlock module to enhance smoke feature extraction and improve detection accuracy with low computational cost. Meanwhile, we design a Lightweight Efficient Convolutional Detection head (LECD), which improves smoke target recognition and localization while reducing the number of parameters and computational overhead of the detection head. We also adopt the Minimum Point Distance Intersection over Union (MPDIoU) as the bounding-box regression loss function to improve the localization accuracy of smoke bounding-box regression. In addition, we construct a UAV-perspective Forest Fire Smoke (UFFS) dataset, which contains typical forest fire smoke, nearby thin smoke, distant small-scale smoke, and smoke under diverse background conditions. Experiments were conducted on both the UFFS dataset and the Wildfire Smoke V1 dataset. The experimental results show that, compared with the baseline model, the proposed model reduces the number of parameters by 36.7% and GFLOPs by 42.3% on the UFFS dataset, while improving mAP50 by 1.3% and mAP50-95 by 3.7%. In addition, recall increases by 3.2% and precision increases by 3.6%, indicating an improved trade-off between detection accuracy and model complexity.

Continual learning (CL) aims to enable networks to learn continuously from sequentially arriving task streams while avoiding catastrophic forgetting (CF) of previously learned tasks. In recent years, Orthogonal gradient projection (OGP)-based CL methods have garnered significant attention from the research community due to their remarkable performance. However, existing OGP approaches overlook two critical issues: (1) representation matrices are typically constructed via random sampling, which introduces misclassified and class-imbalanced samples into the projection basis, contaminating important gradient directions and degrading stability; and (2) task-specific output scale variations induce domain drift, resulting in projection bias that weakens orthogonal constraints across tasks. To address these limitations, we propose Stable Gradient Projection for Continual Learning (STEP), a plug-and-play enhancement framework for OGP-based CL that integrates Correctness-aware Balanced Sampling (CBS) to construct purified and class-balanced projection subspaces using only correctly classified samples, and Sigmoid Attention Constraint (SAC) to enforce consistent output scaling via a sigmoid-based gating mechanism, thereby mitigating scale-induced projection bias. Extensive experiments on Split CIFAR-100, CIFAR-100 Superclass, and 5-Datasets demonstrate that STEP consistently improves state-of-the-art OGP methods, achieving up to +1.4% average accuracy (ACC) gains on Split CIFAR-100, improving backward transfer (BWT) from − 0.37 to − 0.09 for GPM and from − 1.06 to − 0.73 for SGP, and attaining 93.28% ACC with positive BWT (0.17) on 5-Datasets. These results validate STEP as a simple yet effective strategy for enhancing stability–plasticity balance in OGP-based CL.

Occluded person re-identification (Re-ID) is a challenging task, as non-target pedestrians or surrounding obstacles often interfere with the visual cues of the target person, making it difficult for models to effectively learn discriminative feature representations. Most existing methods focus on salient body parts via spatial partitioning or external cues; however, they are either limited in capturing diverse semantic information or tend to introduce additional network complexity. To address these issues, we propose a Diversity Feature Learning Network (DFLNet). Specifically, a Scene-Level Occlusion (SLO) strategy is designed to automatically simulate two common occlusion scenarios by modeling the relative spatial relationships between the target person and surrounding occluders in real-world scenes. Subsequently, multiple class tokens are introduced to capture diverse representations of the target identity. A Token Diversity Constraint (TDC) loss is further imposed on these class tokens to encourage the learning of discriminative and diverse feature embeddings. Finally, we design a Diversity Feature Fusion (DFF) module, which facilitates the interaction and integration of dual-branch features by modeling global feature correlations and optimizing inter-feature distribution distances. Extensive experiments on occluded, partial, and holistic Re-ID datasets demonstrate the effectiveness of the proposed DFLNet.

Unmanned aerial vehicle (UAV) multi-object tracking (MOT) is challenging because small targets produce unstable confidence and localization, while viewpoint changes and short occlusions make association ambiguous. We present OATrack, an integrated association framework tailored to UAV small-object tracking. OATrack combines detection–track pair quality based on detection reliability and current-state geometric consistency, quality-gated track-state protection with a track-specific scale prototype, and three-stage progressive assignment for active-track matching and recently lost-track recovery. A standardized YOLO11m-smallobj detection cache supplies identical detections to all association methods and enables reproducible fixed-input evaluation. On VisDrone-MOT val, OATrack reaches HOTA 38.5847 versus 38.4797 for official Ultralytics ByteTrack while reducing identity switches (IDSW) by 69.2%, track fragmentation by 29.7%, and false positives by 68.8%. On the complete 20-sequence UAVDT test split, OATrack improves HOTA, MOTA, and IDF1 from 41.4345, 7.5162, and 51.4916 to 42.7743, 24.8760, and 53.2829, respectively, and reduces IDSW from 639 to 295. HOTA-component and size/occlusion analyses further demonstrate stronger association consistency and false-positive control across the two UAV benchmarks.