Fusion-based hyperspectral image super-resolution aims to fuse low-resolution hyperspectral images (LR-HSIs) and high-resolution multispectral images (HR-MSIs) to reconstruct high spatial and high spectral resolution images. Current methods typically apply direct fusion from the two modalities without effective supervision, leading to an incomplete perception of deep modality-complementary information and a limited understanding of inter-modality correlations. To address these issues, we propose a simple yet effective solution for unsupervised HMIF, revealing that modality decoupling is key to improving fusion performance. Specifically, we propose an end-to-end self-supervised Modality-Decoupled Spatial-Spectral Fusion (MossFuse) framework that decouples shared and complementary information across modalities and aggregates a concise representation of both LR-HSIs and HR-MSIs to reduce modality redundancy. Also, we introduce the subspace clustering loss as a clear guide to decouple modality-shared features from modality-complementary ones. Systematic experiments over multiple datasets demonstrate that our simple and effective approach consistently outperforms the existing HMIF methods while requiring considerably fewer parameters with reduced inference time. The source code is in MossFuse .

Hyperspectral image change detection (HSI-CD) suffers from severe annotation scarcity and complex change patterns, which fundamentally limit the effectiveness of directly fine-tuning pre-trained foundation models. Although semi-supervised learning provides a promising direction, existing approaches mainly rely on confidence-based pseudo-label selection, leading to limited data diversity or severe error propagation. In this paper, we propose Influence-Aware Semi-supervised Fine-tuning (IA-SFT), a novel framework that evaluates the influence of pseudo-labels on model decision behavior to identify truly valuable supervision signals. Instead of confidence-based selection, IA-SFT evaluates each low-confidence pseudo-label by measuring its impact on labeled data, enabling reliable filtering of high-value pseudo-labels with minimal noise. To further adapt foundation models to HSI-CD, we design an Adaptive Fusion Change Decoder (AFCD) that jointly models global semantic consistency and local change details. Extensive experiments on three benchmark datasets demonstrate that IA-SFT consistently improves pseudo-label quality and detection performance, achieving superior accuracy compared to state-of-the-art methods. Additional analyses validate the transferability of IA-SFT when integrated into different frameworks in a plug-and-play manner. Code will be released.

Open-vocabulary object detection (OVOD) has shown significant potential in real-world remote sensing applications, thanks to its adaptability to open category spaces. Existing OVOD methods mainly rely on pretrained vision-language models (VLMs) to recognize unknown categories. However, VLMs are usually trained on image-text pairs. Whilst possessing strong global semantic modeling capabilities, they struggle to accurately capture the relationship between object instances and their precise locations. To address this problem, this article presents LILK, which learns instance-level knowledge from image-level data through weak supervision. Specifically, an image-level weakly supervised knowledge injection (IWKI) module is first created, which represents instance-level semantic information by introducing global classification queries, while aligning image-level supervision with instance-level supervision through a distinct query promotion (DQP) strategy. Second, a quality-aware pseudo-label rectification (QPLR) module is developed, which filters candidate boxes using dual thresholds based on detector confidence and semantic consistency, while incorporating a SAM2-based relocation mechanism to enhance spatial localization accuracy of pseudo labels. Finally, an image-guided query enhancement (IGQE) module is introduced to guide and reinforce detection queries using instance-level image features, providing the detector with stable and modality-consistent visual References. Extensive experimental results on three commonly used remote sensing object detection datasets, DIOR, DOTA, and NWPU VHR-10, demonstrate the efficacy of LILK in performing OVOD tasks.

Accurate network topology inference is crucial for analyzing non-cooperative networks, where structural and routing information is unavailable. To address this challenge, we propose CP-AGN, a hybrid framework that combines a causal prior with a lightweight neural refinement module. Specifically, a multivariate Hawkes process is employed to capture potential causal relationships from network event sequences and construct a prior. Building upon this prior, a lightweight convolutional module performs edge pruning and confidence reweighting using pairwise correlation statistics. Extensive experiments on NS-3 simulated datasets under standard TCP/IP protocol assumptions demonstrate that CP-AGN effectively recovers hidden network structures while maintaining robust performance under noisy observations and packet loss.

Recently, diffusion models have brought novel insights to pan-sharpening and notably boosted fusion precision. However, most existing models perform diffusion in the pixel space and train distinct models for different multispectral (MS) sensors, suffering from high inference latency and sensor-specific limitations. In this paper, we present CC-Pan, a cross-sensor latent diffusion framework for efficient pan-sharpening. Specifically, CC-Pan trains a band-wise single-channel variational autoencoder (VAE) to encode high-resolution multispectral (HRMS) images into compact latent representations, naturally supporting MS images with varying band counts across different sensors and establishing a basis for inference acceleration. Spectral physical properties, along with PAN and MS images, are then injected into the diffusion backbone through carefully designed unidirectional and bidirectional interactive control structures, achieving high-precision spatial–spectral fusion in the latent diffusion process. Furthermore, a lightweight region-based cross-band attention (RCBA) module is incorporated at the central layer of the diffusion model, reinforcing inter-band spectral connections to boost spectral consistency and further elevate fusion precision. Extensive experimental results on GaoFen-2, QuickBird, and WorldView-3 demonstrate that CC-Pan outperforms state-of-the-art diffusion-based methods across all three benchmarks, attains a 2–3× inference speedup, and exhibits robust cross-sensor generalization capability on the held-out WorldView-2 sensor without any sensor-specific retraining.

Image generation technology can synthesize condition-specific images to supplement real-world industrial anomaly data and enhance anomaly detection model performance. Existing generation techniques rarely account for the pose and orientation of industrial components in assembly, making the generated images difficult to utilize for downstream application. To solve this, we propose a novel image synthesis approach, called PostureObjectStitch, that achieves accurate generation to meet the requirement of industrial assembly. A condition decoupling approach is introduced to separate input multi-view images into high-frequency, texture, and RGB features. The feature temporal modulation mechanism adapts these features across diffusion model time-steps, enabling progressive generation from coarse to fine details while maintaining consistency. To ensure semantic accuracy, we introduce a conditional loss that enhances critical industrial elements and a geometric prior that guides component positioning for correct assembly relationships. Comprehensive experimental results on the MureCom dataset, our newly contributed DreamAssembly dataset, and the downstream application validate the outstanding performance of our method.

Multimodal Large Language Models (MLLMs) based agents have demonstrated remarkable potential in autonomous web navigation. However, handling long-horizon tasks remains a critical bottleneck. Prevailing strategies often rely heavily on extensive data collection and model training, yet still struggle with high computational costs and insufficient reasoning capabilities when facing complex, long-horizon scenarios. To address this, we propose M^2, a training-free, memory-augmented framework designed to optimize context efficiency and decision-making robustness. Our approach incorporates a dual-tier memory mechanism that synergizes Dynamic Trajectory Summarization (Internal Memory) to compress verbose interaction history into concise state updates, and Insight Retrieval Augmentation (External Memory) to guide the agent with actionable guidelines retrieved from an offline insight bank. Extensive evaluations across WebVoyager and OnlineMind2Web demonstrate that M^2 consistently surpasses baselines, yielding up to a 19.6

Open-vocabulary object detection aims to localize and recognize objects from an open category space without being restricted to predefined classes. While recent training-free methods leverage powerful pretrained foundation models to avoid costly detector training, their performance in remote sensing scenarios remains limited due to category hallucination, insufficient category awareness, and unreliable candidate predictions. To address these challenges, CMBI is proposed as a Cross-Modal Bayesian Inference for training-free open-vocabulary object detection in remote sensing images as a rigorous Bayesian Maximum A Posteriori (MAP) estimation problem. Integrating SAM3 as the base detector and Qwen3 as the semantic assistant, CMBI performs collaborative reasoning through three probabilistically grounded components. First, a global semantic prior formulation mechanism is introduced to tightly constrain the open hypothesis space, effectively suppressing category hallucinations. Second, an adaptive conditioned likelihood estimation strategy employs dynamic visual anchors alongside text to sharpen spatial probability distributions, enhancing the perception of hard-to-recognize targets. Third, a posterior calibration and MAP decision module extracts local semantic evidence to iteratively calibrate intermediate probabilities, thereby eliminating spatial grouping ambiguities and facilitating more robust, bounding box predictions. Extensive experiments on three representative remote sensing benchmarks, DIOR, NWPU VHR-10, and HRRSD, demonstrate that CMBI consistently achieves superior performance over existing training-free baselines. These results validate the effectiveness of cross-modal collaboration for improving the accuracy and robustness of training-free open-vocabulary object detection in remote sensing imagery.

Semi-supervised joint classification of Hyperspectral Images (HSIs) and LiDAR-derived Digital Surface Models (DSMs) remains challenging due to scarcity of labeled pixels, strong intra-class variability, and the heterogeneous nature of spectral and elevation features. In this work, we propose a Hybrid Mamba–Graph Fusion Network (HMGF-Net) with Multi-Stage Pseudo-Label Refinement (MS-PLR) for semi-supervised hyperspectral–LiDAR classification. The framework employs a spectral–spatial HSI backbone combining 3D–2D convolutions, a compact LiDAR CNN encoder, Mamba-style state-space sequence blocks for long-range spectral and cross-modal dependency modeling, and a graph fusion module that propagates information over a heterogeneous pixel graph. Semi-supervised learning is realized via a three-stage pseudolabeling pipeline that progressively filters, smooths, and re-weights pseudolabels based on prediction confidence, spatial–spectral consistency, and graph neighborhood agreement. We validate HMGF-Net on three benchmark hyperspectral–LiDAR datasets. Compared with a set of eight state-of-the-art (SOTA) baselines, including 3D-CNNs, SSRN, HybridSN, transformer-based models such as SpectralFormer, multimodal CNN–GCN fusion networks, and recent semi-supervised methods, the proposed approach delivers consistent gains in overall accuracy, average accuracy, and Cohen’s kappa, especially in low-label regimes (10% labeled pixels). The results highlight that the synergy between sequence modeling and graph reasoning in combination with carefully designed pseudolabel refinement is essential to maximizing the benefit of abundant unlabeled samples in multimodal remote sensing scenarios.

Gradient compression can reduce communication overhead. However, current static sparsity techniques may disturb gradient dynamics, resulting in unstable model convergence and reduced feature discriminative ability, whereas transmitting the complete gradient leads to high costs. To address this issue, inspired by nonequilibrium thermodynamics, this paper proposes a Physics-guided Gradient Sparsification Criterion (PGSC). Specifically, we formulate a continuous field equation based on the gradient magnitude distribution, deriving an adaptive decay rule for the sparsification threshold during the training phase. We then dynamically adjust the sparsification threshold according to this rule, effectively addressing the complexity of multimodal features and ensuring consistent information transmission. Our method achieves adaptive co-optimization of gradient compression and model accuracy by establishing a dynamic equilibrium mechanism between gradient dissipation and information entropy. This approach ensures stable convergence rates while preserving the gradient structure of multi-scale features. Extensive experiments on public datasets, including CIFAR-10, MNIST, and FLIR_ADAS_v2, demonstrate significant advantages over competitors such as TopK and quantization compression, while also reducing communication costs.

Diffusion models have shown remarkable capabilities for arbitrary style transfer. However, existing methods remain limited in their ability to balance content preservation with style integration. To address these issues, we propose TransferAnything, a unified framework that introduces two self-attention-driven constraints in the latent space to jointly enforce style alignment and content fidelity. To mitigate gradient conflicts that cause style-content leakage or insufficient stylization, we introduce two frequency-aware modulation strategies grounded in Fourier analysis. Specifically, the Content-Focused Frequency Calibration (CFFC) strategy suppresses low-frequency components in the initial latent to guide optimization toward structural content preservation, the Amplitude-Guided Prior (AGP) strategy injects the amplitude of the style image into the latent image as a prior, effectively enhancing stylistic expressiveness. By modulating frequency components within the latent representation, our method provides fine-grained control over stylization. Extensive experiments demonstrate that TransferAnything surpasses state-of-the-art methods in structural fidelity and stylistic coherence. Code is available at https://github.com/HuanBor/TransferAnything

Corner rounding in lithography describes the unintended transformation of designed sharp corners into curved features during pattern transfer. This smoothing effect is primarily driven by light diffraction and is further influenced by factors including mask design, numerical aperture, and process parameters. As a critical two-dimensional effect, inadequate control of corner rounding compromises pattern fidelity directly and ultimately degrades the performance and reliability of microelectronic devices. The adoption of extreme ultraviolet (EUV) lithography, which enables continued scaling of feature sizes and device density, imposes even more stringent requirements on corner rounding control. The reflective optics inherent to EUV systems, combined with the shorter exposure wavelength, introduce additional complexities in preserving corner acuity. This paper presents a comprehensive study of the mechanisms governing corner rounding in EUV lithography, using rigorous imaging simulations to dissect and quantify the individual effects of projection optics, mask diffraction, and resist behavior. By leveraging aerial image simulations and computational modeling, we establish the fundamental limits of corner sharpness and propose mitigation strategies to achieve minimal corner radius in both 0.33 NA and 0.55 NA EUV systems. Our results demonstrate notable improvements in corner sharpness and overall image quality, offering an actionable methodology for enhancing two-dimensional pattern fidelity in advanced semiconductor manufacturing.

Ensuring structural robustness is a fundamental goal in complex network analysis and control. Existing approaches primarily focus on either local connectivity or information diffusion, often neglecting the critical roles of nodes in preserving overall network integrity. To address this limitation, we propose a hybrid centrality method (HCM) integrating local and global network information to quantify node importance. Specifically, we define local dispersion centrality by combining node degree and local clustering coefficient to capture the dispersion of a node’s neighborhood, and employ betweenness centrality to reflect its global structural significance as bridge nodes. HCM is formulated as a weighted combination of these two measures, with a tunable parameter balancing local and global contributions. It comprehensively assesses node importance and effectively identifies nodes whose removal fragments the network. Extensive experiments on synthetic and real-world networks demonstrate HCM outperforms baselines in network dismantling, with pronounced effectiveness in high-clustering networks.

Computer-aided diagnosis (CAD) technology has become an integral part of early breast cancer diagnosis in medical ultrasound imaging. Nonetheless, the accurate segmentation and classification of breast tumor images continue to pose significant challenges due to the uneven intensity distribution, indistinct boundaries, and the irregular shapes of tumors. A primary reason is that many CAD methods fail to capitalize on the interrelation between tumor segmentation and classification. Furthermore, most existing methods focus on classifying global images or tumor regions of interest (ROIs), often disregarding the interplay between global and local features. This study proposes a segmentation knowledge based global-local attention classification network (SGLA-Net). First, the segment anything model (SAM) is utilized to obtain high-quality segmentation masks from a limited number of annotated samples. Then, global and local feature representations are derived from enhanced images obtained through the segmentation mask. Moreover, a global-local feature interaction (GLFI) block is designed to adaptively integrate global and local information for classification. The proposed method achieved segmentation Dice coefficients of 81.239 % and 80.516 % on the internal and external datasets, respectively. In terms of classification, the method obtained Area Under the Curve (AUC) values of 0.9532 and 0.8521 on the internal and external datasets, surpassing five state-of-the-art breast ultrasound classification methods.

Single-Pixel Imaging, with its unique imaging mechanism, offers a novel approach to edge detection in complex environments. However, existing methods mainly optimize individual stages and still lack a unified strategy for holistic optimization. To address this, this paper proposes a Frequency-Dependent Framework for Directional Edge Detection in Single-Pixel Imaging (FDDE-SPI), which uses the frequency-domain properties of edges as a global design criterion to achieve end-to-end collaborative optimization. In the modulation stage, sampling and filtering work in close coordination: while selecting key frequencies based on spectral energy distribution, the system dynamically matches the optimal smoothing scale to the scale characteristics of edges. In the reconstruction stage, the system enhances the robustness of two-step phase-shifting acquisition and employs a directional total variation regularization algorithm aligned with the filtering direction for efficient edge extraction. Simulation and experimental results demonstrate that this framework outperforms representative existing methods in terms of both SNR and F1-score, indicating its potential for robust edge detection in challenging optical environments.

Existing diffusion-based object removal and inpainting methods often fail to recover the fine structural and textural details of small objects. This is primarily due to the VAE encoder’s downsampling, which inevitably compresses small masked regions and causes significant detail loss, while the decoder’s upsampling alone cannot fully restore the lost fine details. However, the adverse effects of this fixed compression can be mitigated by enlarging the perspective of these regions. To this end, we propose ReFocusEraser, a two-stage framework for small object removal that combines camera-adaptive zoom-in inpainting with robust context- and shadow-aware repair. In Stage I, a camera-adaptive refocus mechanism magnifies masked regions, and a LoRA-tuned diffusion model ensures precise semantic alignment for accurate reconstruction. However, reintegrating these magnified inpainted regions into the original image introduces challenges due to VAE asymmetry, such as color shifts and seams. Stage II addresses these issues by fine-tuning an additional decoder to create a seam- and shadow-aware module that eliminates residual artifacts while preserving background consistency. Extensive experiments demonstrate that our proposed RefocusEraser achieves state-of-the-art performance, outperforming existing methods across benchmark datasets.

Hyperspectral (HS) pansharpening is a key preprocessing step in various remote sensing applications, which produces a high-resolution (HR) HS image through the fusion of a low-resolution HS image with an HR panchromatic (PAN) image. Recent deep learning-based methods, particularly those employing convolutional neural networks and Transformers, have achieved impressive progress owing to their strong representational capacity. However, the fusion performance of these methods may degrade in the absence of large-scale datasets. Moreover, the attention mechanism in Transformers results in quadratic computational complexity. To alleviate these limitations, we propose MPSRNet-Diff, a two-stage fusion framework that avoids directly learning the relationship between observed images and ideal HR-HS images. In the first stage, a prior HS image is generated using a pretrained diffusion model, which benefits from strong generalization capabilities learned from large-scale datasets. To improve the efficiency of generating this prior HS image, the low-rank property of HS data is exploited, and a principal component analysis-based band selection strategy is proposed to reduce spectral dimensionality. This prior HS image serves as a PAN-like proxy, guiding the HR-HS reconstruction by regularizing the solution space and improving the generalization ability to unseen data. In the second stage, a Mamba-based progressive superresolution network (MPSRNet) is proposed to enhance the spatial details of the HS image in a step-by-step manner to learn global dependencies without quadratic complexity. At each refinement step, the MPSRNet receives the prior HS image, PAN image, and HS image to be superresolved as inputs. To fully exploit both intrainput and interinput global dependencies, a mutual-guided Mamba block is introduced, which leverages state space models for efficient global information modeling. Extensive experiments on four datasets demonstrate that MPSRNet-Diff is competitive with state-of-the-art techniques in both quantitative metrics and visual quality, validating its effectiveness and robustness.