Landslide detection and segmentation are critical for disaster risk assessment and management. However, achieving accurate segmentation remains challenging due to the complex nature of landslide terrains and the limited availability of high-quality labeled datasets. This paper proposes an enhanced U-Net++ model for semantic segmentation of landslides in the Wenchuan region using the CAS Landslide Dataset. The proposed model integrates multi-scale feature extraction and attention mechanisms to enhance segmentation accuracy and robustness. The experimental results demonstrate that ASK-UNet++ outperforms traditional methods, achieving a mean intersection over union (mIoU) of 97.53%, a Dice coefficient of 98.27%, and an overall accuracy of 96.04%. These findings highlight the potential of the proposed approach for improving landslide monitoring and disaster response strategies.

Unmanned marine vehicles (UMVs) are one of the most crucial components of the underwater communication and surveillance system in the marine and oceanographic environment. This research introduces a novel approach to stabilize the connection in an unstable underwater environment by inheriting features of communication velocities. The innovative proposed technique effectively divides the entire communication underwater by distributing the burden and ensuring trouble‐free and efficient communication and surveillance. In existing approaches, troubleshooting communication issues often results in delays or slow performance. However, the proposed method focuses on establishing an efficient communication link to increase the throughput of the deployed underwater network. To achieve this, we have implemented two different communication channels, acoustic and wireless optical, that provide redundant pathways for communication, which enhances reliability and resilience. In the event of a denial‐of‐service attack or failure in one communication channel, the other channel becomes the primary means of communication, ensuring uninterrupted data transmission. We have used an attack prediction machine‐learning model to further improve system security. Random forest (RF), naïve Bayes classifiers, and machine learning methods like support vector machines (SVM), multilayer perceptrons (MLP), and k‐nearest neighbors (KNN) are used to forecast the likelihood of an attack. To stop unwanted access or malicious activity, the associated communication channel is shut when a model shows a high probability of an attack. The efficiency of our recommended control strategy is confirmed by a simulation study of the proposed networked UMVs system. The suggested system delivers improved robustness and secure communication for real‐time situations in the underwater environment by using machine learning‐based attack prediction, redundant communication channels, and varied velocities to stabilize the connection. SVM achieves high accuracy at 98%. MLP is not far behind, with a 96% accuracy rate. The higher accuracy of SVM and MLP in predicting DoS assaults in UMVs is highlighted by the lesser accuracies of 66.3.

Landslide hazard assessment increasingly demands the joint analysis of heterogeneous remote sensing data; however, automating this process remains difficult due to the pronounced resolution and texture discrepancies existing between satellite and aerial sensors. To address these limitations, this study proposes a robust segmentation framework capable of extracting sensor-robust representations. The framework leverages a DINOv3 transformer encoder and exploits representations from multiple transformer layers to capture complementary visual information, ranging from fine-grained surface textures to global semantic contexts, overcoming the receptive field constraints of conventional CNNs. Experiments on the Longxi satellite dataset achieve a Dice coefficient of 0.96 and an IoU of 0.938, and experiments on the Longxi UAV dataset achieve a Dice coefficient of 0.965 and an IoU of 0.941. These results show consistent segmentation performance on both the Longxi satellite and UAV datasets, despite differences in spatial resolution and surface appearance between acquisition platforms.

Semantic segmentation of very-high-resolution remote sensing imagery benefits from combining RGB appearance with Digital Surface Model (DSM) height information, especially in urban scenes where spectrally similar objects often differ in elevation. On the ISPRS Vaihingen and Potsdam benchmarks, BATFNet achieves mIoU scores of 84.06% and 85.31%, respectively, outperforming representative RGB–DSM fusion baselines on most land-cover categories. BATFNet is a supervised boundary-aware Transformer fusion network that uses DSM-derived edge priors to guide bidirectional cross-modal attention and decoder refinement. With a dual-branch ResNet-50 backbone for modality-specific feature extraction, the proposed framework effectively integrates RGB and DSM information while recovering fine spatial details. These results show that exploiting DSM-derived structural cues improves boundary delineation and reduces confusion among spectrally similar urban classes.

High-risk landslide extraction from remote sensing imagery is a fundamental task for geological disaster prevention, emergency response, and land-use planning in mountainous regions. Although deep-learning semantic segmentation has substantially advanced landslide detection from optical imagery, existing methods still suffer from three critical limitations: (i) Interferometric Synthetic Aperture Radar (InSAR) deformation data are treated as auxiliary channels and dominated by optical features during fusion; (ii) predicted masks exhibit fragmented boundaries and incomplete delineation due to the absence of deformation continuity constraints reflecting the physical coherence of slope movements; and (iii) heavy Transformer backbones hinder practical deployment over large areas. To address these issues, we propose a Deformation-Integrity-Guided Symmetric Fusion Network (DIGSFNet) for high-risk landslide extraction from InSAR and optical imagery. The framework consists of three components: a Symmetric Deformation-Aware Encoder (SDAE) that treats InSAR and optical modalities as equal information sources through modality-aware adapters and dynamic sparse cross-modal fusion; a Deformation Integrity Prior Decoder (DIPD) that imposes deformation continuity and boundary-gradient consistency as physical priors to enforce mask completeness and boundary accuracy; and a Lightweight Deployable Student Network (LDSN) obtained via cross-modal knowledge distillation and INT8 quantization for efficient inference. Experiments on the Nanning High-Hazard Landslide Segmentation (Nanning-HHLS) dataset and the public HAEFNet benchmark covering the Qinghai–Tibet–Sichuan landslide-prone regions show that the full DIGSFNet achieves state-of-the-art extraction accuracy, reaching 83.57% and 78.92% mIoU on the two datasets and surpassing the strongest competing method by 2.63 and 3.68 percentage points with a Recall of 91.48% on Nanning-HHLS, while its distilled lightweight student retains 79.24% mIoU at 218 frames per second after INT8 quantization, enabling efficient large-area operational deployment.

Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities across a diverse range of multimodal tasks. However, these models suffer from a core problem known as text dominance: they depend heavily on text for their inference, while underutilizing other modalities. While prior work has acknowledged this phenomenon in vision-language tasks, often attributing it to data biases or model architectures. In this paper, we conduct the first systematic investigation of text dominance across diverse data modalities, including images, videos, audio, time-series, and graphs. To measure this imbalance, we propose two evaluation metrics: the Modality Dominance Index (MDI) and the Attention Efficiency Index (AEI). Our comprehensive analysis reveals that text dominance is both significant and pervasive across all tested modalities. Our in-depth analysis identifies three underlying causes: attention dilution from severe token redundancy in non-textual modalities, the influence of fusion architecture design, and task formulations that implicitly favor textual inputs. Furthermore, we propose a simple token compression method that effectively rebalances model attention. Applying this method to LLaVA-7B, for instance, drastically reduces its MDI from 10.23 to a well-balanced value of 0.86. Our analysis and methodological framework offer a foundation for the development of more equitable and comprehensive multimodal language models.