Accurate localization of underwater sensor nodes is a fundamental yet challenging problem in underwater wireless sensor networks. Localization performance is severely affected by complex environmental factors, including measurement noise, Non-Line-Of-Sight (NLOS) propagation, absorption loss, and path attenuation. To address these challenges, this paper proposes an efficient and scalable localization mechanism based on Received Signal Strength Difference (RSSD). First, the Expectation–Maximization (EM) algorithm is applied to identify NLOS links and estimate their corresponding biases. Then, a weighted Huber loss function combined with an adaptive weighting strategy is introduced to construct an Adaptive Weighted Loss Function, which suppresses noise, adaptively reduces the influence of NLOS paths, and compensates residual biases. Based on this formulation, the localization task is transformed into an optimization problem. Furthermore, an improved Grey Wolf Optimizer (IGWO) incorporating an Adaptive Convergence Factor (ACF) and an Elite-Guided (EG) strategy is developed to enhance global search capability and convergence stability. Extensive simulation results demonstrate that the proposed method achieves significantly higher localization accuracy than existing approaches.

Underwater images typically suffer from color distortion and reduced clarity due to wavelength-dependent light absorption and scattering. Many existing learning-based methods learn a direct pixel-domain mapping from degraded images to enhanced results, while seldom exploiting explicit frequency-domain priors that are important for high-quality restoration. To address this, we propose DPF-Net, a Dual-Path Frequency-domain Fusion Network. Specifically, we first decompose the input into low- and high-frequency components via the discrete wavelet transform (DWT). We then design a low-frequency branch integrating Low-Frequency Guided Attention (LFGA) and the Color Feature Extraction Module (CFEM), and a high-frequency branch integrating High-Frequency Aware Modulation (HFAM) and the Detail Feature Extraction Module (DFEM), to capture frequency-specific characteristics. During upsampling, we introduce a WaveUp module consisting of the Dual-path Up-Interaction Module (DUIM) and a Multi-kernel Convolution-based Fusion Network (MCFN) to enable effective cross-frequency feature interaction and fusion, producing high-quality enhanced images. Extensive experiments on benchmark datasets demonstrate that DPF-Net achieves superior qualitative and quantitative performance compared with state-of-the-art methods, and real-world tests further verify its practical potential. Code is available at https://github.com/YuMa-star-lab/DPF-Net.

Fountain codes with online adaptation (OFCs) are promising for underwater acoustic sensor networks (UASNs), since they exploit limited feedback to reduce transmission over head. Yet, the performance of conventional OFCs is severely degraded in UASNs due to high error rates, long propagation delays, and sparse feedback, resulting in poor recovery efficiency and high energy cost. To address these challenges, we propose an energy-aware dual-perception OFC framework driven by deep reinforcement learning (DRL-OFC). In this design, a DRL agent jointly perceives channel dynamics and feedback sparsity to learn an optimal degree distribution that suppresses redundant coding and enhances intermediate decoding. In addition, a feedback aware transmission strategy is developed to cope with the long delay characteristics of UASNs, further reducing unnecessary retransmissions. Simulation results show that DRL-OFC achieves significant gains over existing OFC schemes in terms of decoding efficiency, transmission overhead, recovery performance, and energy consumption, confirming its suitability for resource constrained underwater communications.

Factual consistency is an important quality in dialogue summarization. Large language model (LLM)-based automatic text summarization models generate more factually consistent summaries compared to those by smaller pretrained language models, but they face deployment challenges in real-world applications due to privacy or resource constraints. In this paper, we investigate the use of symbolic knowledge distillation to improve the factual consistency of smaller pretrained models for dialogue summarization. We employ zero-shot learning to extract symbolic knowledge from LLMs, generating both factually consistent (positive) and inconsistent (negative) summaries. We then apply two contrastive learning objectives on these summaries to enhance smaller summarization models. Experiments with BART, PEGASUS, and Flan-T5 indicate that our approach surpasses strong baselines that rely on complex data augmentation strategies. Our approach demonstrates improved factual consistency while preserving coherence, fluency, and relevance, as verified by both automatic evaluation metrics and human assessments. We provide access to the data and code to facilitate future research.

With the increasing deployment of Underwater Acoustic Sensor Networks (UASNs) in various marine applications, ensuring network security has become a significant concern. Trust models offer an adaptive security mechanism for such networks. This paper proposes a dual-layer trust model (DLtrust) based on digital twin (DT) architecture. First, a network framework is constructed using DTs, in which the Local DT employs an adaptive long short-term memory (LSTM) algorithm to detect conventional attacks, providing the primary defense mechanism. DLtrust further interacts with the Cloud DT in real time to obtain the optimal weighting of trust evidence, thereby accelerating model convergence. Additionally, a reinforcement learning scheme is integrated into the Cloud DT to optimize dynamic trust evaluation and refine the evidence aggregation strategy of the Local DT. By continuously perceiving the global network status and adjusting trust parameters, the Cloud DT effectively guides the Local DT to detect anomalous behavior, especially in cases of compromised models. To enhance resilience against Byzantine attacks, the model incorporates the multi-Krum algorithm, strengthening advanced defense capabilities. Simulation results demonstrate that DLtrust outperforms existing trust models in terms of average detection accuracy and error rate.

Underwater Wireless Sensor Networks (UWSNs) have attracted considerable attention due to their critical roles in ocean data acquisition, marine exploration, and disaster prevention. As diverse underwater applications continue to emerge, there has been a significant increase in both the volume and variety of sensed data. This necessitates reliable and efficient data transmission to data centers, particularly given the challenging underwater environment. To address these demands, this paper introduces EVAOR: an Efficient and Void-Avoidable Opportunistic Routing Protocol based on the Genetic Algorithm (GA). EVAOR incorporates four key strategies to optimize routing performance. First, it implements an Enhanced Local Topology Awareness Mechanism to generate a high-quality initial population, accelerating the convergence of the GA. Second, a novel crossover scheme and scaling function are employed to enhance the GA’s search efficiency. Third, a multi-factor fitness function is designed to leverage local topological information, ensuring high-quality chromosome generation. Finally, a multiarmed bandit-based depth adjustment algorithm is proposed to optimize void node positioning, effectively restoring nodes from void zones and improving overall network performance. Experimental results demonstrate that EVAOR surpasses existing routing protocols in terms of packet delivery ratio and end-to-end delay, with the depth adjustment algorithm showing superior efficiency compared to traditional depth adjustment techniques.

Online Fountain Codes (OFCs) adapt their encoding strategy to decoder feedback, making them attractive for Underwater Acoustic Sensor Networks (UASNs) with stringent delay and energy constraints. Yet, conventional OFCs perform poorly under the harsh conditions of UASNs, where high error rates, long propagation delays, and limited feedback severely restrict efficiency. This work identifies two key challenges: balancing recovery efficiency with intermediate decoding, and reducing encoding and decoding complexity under sparse feedback. To address them, we propose a Dual-Perception Fountain Code framework based on Deep Reinforcement Learning (DRL-OFC). A DRL agent learns optimized degree distributions to reduce redundant packets and improve decoding, while a feedback-aware strategy adapts transmissions to UASNs conditions. Simulations show that DRL-OFC achieves lower overhead, stronger intermediate recovery, fewer coded packets, and better energy efficiency than existing OFC schemes, confirming its suitability for resource-constrained underwater networks.

To address the pressing demands of marine resource exploration, this paper investigates the problem of node localization in underwater acoustic wireless networks and proposes a two-stage progressive collaborative optimization framework to overcome the performance limitations of the traditional DV-Hop algorithm. Error sources of DV-Hop in underwater scenarios are systematically analyzed, and a quantitative localization model, NSGA-II-DV-Hop, is established to incorporate both hop count estimation errors and position calculation errors. Based on this analysis, a two-stage optimization strategy is developed. In the first stage, an adaptive particle swarm optimization model, PSO-DV-Hop, is designed, where a dynamic inertia weight adjustment mechanism enhances global search efficiency, thereby reducing localization error and energy consumption. In the second stage, a Pareto-Based evolutionary optimization model, NSGA-II-DV-Hop, is introduced to realize simultaneous optimization of localization accuracy and energy efficiency through Pareto frontier analysis. Experimental results demonstrate that PSO-DV-Hop improves localization performance by reducing the average error by 18.2% and lowering the number of convergence iterations by 61%. Building on this, NSGA-II-DV-Hop further extends the network lifetime by 26.9%, reduces average node energy consumption by 10.07%, and achieves an additional 8.67% energy optimization.

Marine mammals are crucial subjects for marine audio studies, particularly in the development of retrieval algorithms for encrypted marine mammal audio.To address the issues of low retrieval performance and high content redundancy, and to ensure security of existing marine mammal audio retrieval algorithms, this paper proposes an encrypted marine mammal audio retrieval algorithm based on biohashing. The server terminal first extracts the sub-band CQT energy-entropy ratio of audio to construct biometric dataset. by querying the key and corresponding classified biometric in the key-address index table constructed based on SVM classification, the key-controlled orthogonal matrix and the corresponding classified biometric generate the biosafety template, and the template is quantified to obtain biohashing. Finally, the dual-threshold audio segmentation and position mapping are adopted to perform hash reconstruction to obtain the hash index. Meanwhile, the Logistic-AES encryption method encrypts audio clips to construct the encrypted audio library. Experimental results demonstrate that sub-band CQT energy-entropy ratio has better robustness and discrimination, which results in the higher precision and recall of the retrieval system. And the proposed audio segmentation algorithm reduces the redundant part of audio, improving the efficiency of retrieval system. Meanwhile, the designed encryption method is sensitive to the key and disrupts audio correlation, which ensures the resistance of encrypted audio to statistical attacks. Furthermore, the biosafety template of biohashing has better recoverability, security and diversity.

In the realm of underwater wireless communication, vast oceanic expanses often demand large-scale deployment of Underwater Wireless Sensor Networks (UWSNs). UWSNs rely on acoustic communication channels, presenting distinct challenges like prolonged propagation delays, restricted bandwidth, and dynamic topologies. Furthermore, the far-reaching and multi-path nature of acoustic signals results in significant hidden terminal problems and ubiquitous interference between neighboring nodes. Therefore, an efficient medium access control (MAC) protocol is crucial for optimizing UWSN performance. This article proposes IC-MAC, a MAC protocol tailored for UWSNs to avoid collisions and improve network performance. IC-MAC employs distributed clustering to group sensor nodes and the cluster head degree is defined for each node, which is a coefficient that accentuates nodes characterized by a higher incidence of collision associations. To identify interfering nodes and construct an interference-free graph, an interference identification algorithm is proposed. In addition, a heuristic graph coloring technique, guided by particle swarm optimization, allocates time slots efficiently to achieve collision-free transmission scheduling and enhanced spatial reuse. Simulations demonstrate the effectiveness of the IC-MAC protocol in enhancing throughput, reducing delay, and improving packet delivery ratio and energy efficiency. This is achieved through efficient spatial resource utilization and robust management of collisions and interference, specifically tailored for underwater acoustic channels, outperforming existing MAC protocols.

Underwater Wireless Sensor Networks (UWSNs) are integral to marine exploration yet confront significant security concerns. In contrast to terrestrial WSNs, underwater acoustic channels are characterized by their constrained bandwidth, considerable propagation delays, and a higher bit error rate. These factors facilitate adversaries’ ability to intercept network transmissions and acquire sensitive data. Furthermore, the constraints of energy resources in UWSNs pose additional challenges in harmonizing privacy preservation with energy conservation. This study introduces a novel Privacy-Preserving Data Fusion Algorithm (DSPDA) tailored for UWSNs. The DSPDA circumvents the shortcomings of conventional privacy algorithms by employing dynamic sharding to minimize transmission demands and augment data fusion precision, which adjusts the size of data shards relative to nodal distances, thereby safeguarding confidentiality. Furthermore, it enhances network-wide energy optimization by aligning the energy consumption of cluster heads with their respective child nodes, thus averting disproportionate energy depletion and potential network dysfunction. Our simulation results demonstrate that the DSPDA algorithm notably diminishes communication overhead by approximately 37% in comparison to the SMART algorithm and delivers a further 20% efficiency improvement over the EEHA algorithm.

Underwater Acoustic Sensor Networks (UASNs) have garnered significant interest over recent decades, yet they continue to confront substantial challenges concerning security and privacy. The inherent openness of acoustic communication in such networks introduces profound risks to node privacy and overall network security. External attackers are capable of retracing data streams to identify the source node, thus jeopardizing the confidentiality of the data origin. Notably, the subtleties of passive attacks in UASNs render them more elusive compared to active attacks. Furthermore, conventional research often fails to reconcile security needs with energy efficiency. Addressing these concerns, this paper concentrates on mitigating passive attacks in UASNs while striving to harmonize security with network performance. We introduce a Source Location Privacy Scheme based on the Multi-Armed Bandit and Pseudo-packet Scheduling for UASNs (MP-SLP). The methodology commences with the sink node executing geographic data acquisition and neighbor node identification using a flood-based technique. Subsequently, through a Multi-Armed Bandit (MAB) framework, the source node designates an intermediate node for data packet transmission. The sink node then employs a location-aware opportunistic routing strategy to establish authentic routes that combine both randomness and reduced energy consumption for the transmission of actual packets. Alongside, branch nodes implement a pseudopacket scheduling tactic aimed at obstructing adversarial efforts to track source locations. Simulation results demonstrate that the proposed scheme proficiently moderates additional energy consumption, maintaining them within acceptable limits, while safeguarding location privacy.

The burgeoning deployment of Underwater Acoustic Sensor Networks (UASNs) for maritime applications highlights the critical need for reliable trust models to defend against internal security threats. Existing trust models are often inadequate due to high packet error rates inherent in underwater communication and a lack of accounting for nodes' traffic behavior. Additionally, conventional UASN architectures suffer from significant latency in gathering and processing trust evidence, which delays the identification of adversarial nodes. Addressing these limitations, this paper proposes the Traffic-Aware and Edge Computing-Enabled Trust Model (TECTM), a solution expressly conceived for UASNs. TECTM integrates environmental models to assess the acoustic environment's impact on communication and employs network traffic analysis as a trustworthy metric for identifying attack patterns. Autonomous Underwater Vehicles (AUVs) serve as edge computing nodes, leveraging a machine learning algorithm to enhance trust assessments within node clusters. Moreover, TECTM introduces a refined trust update mechanism, designed to be responsive to the dynamic underwater environment and complex attack behaviors. Through comparative simulations, TECTM demonstrates enhanced accuracy in the detection of malicious nodes, outperforming other methods.

Underwater Acoustic Sensor Networks (UASNs) have emerged as pivotal contributors to various domains. Never-theless, UASNs grapple with intrinsic challenges, notably propa-gation delays, heightened energy consumption, and inconsistent transmissions, which cumulatively impair the fidelity of under-water communication. Given these challenges, the exigency for a mechanism that assures both energy efficiency and reliable data collection becomes paramount. In this paper, we propose a data acquisition framework with the support of Autonomous Underwater Vehicles (AUVs), utilizing the Hybrid Clustering and Matrix Completion (AHMC) methodology, aiming to enhance the efficiency of data collection in UASNs. Initially, our approach har-nesses a novel hybrid clustering strategy, melding the virtues of the fuzzy c mean (FCM) algorithm with the Firefly Optimization Method (FA) to bolster network performance. The core strategy delineates the formation of energy-optimized clusters through FCM, post which the Elbow method ascertains the optimal K-value. Successively, the FA algorithm becomes instrumental in pinpointing the most suitable cluster head (CH) for each cluster. Furthermore, we introduce an ant colony algorithm tailored for the UASN s, encapsulating factors like energy, transmission latency, and the comprehensive trajectory span of AUV s during their operational phase. This strategic inclusion refines the pheromone update model, seamlessly amalgamating multifaceted elements to chart an optimally efficient AUV pathway. To cul-minate, the intra-cluster data aggregation is honed via a matrix completion methodology. Simulation results attest to the superior performance of our AHMC paradigm, showcasing commendable metrics in energy efficiency, latency, and network lifetime.

The growing significance of marine information in the context of increased human oceanic activities has fostered interest in marine exploration. However, the specific nature of underwater acoustic communication, characterized by propagation delays and fluctuating link quality, presents multifaceted challenges to Internet of Underwater Things (IoUT). The vast openness of the underwater environment further amplifies security vulnerabilities, emphasizing the imperative for secure network routing. This paper introduces GHL-SAR, a fortified routing paradigm to address these challenges. Central to GHL-SAR’s design is its ability to evaluate node trustworthiness based on energy, communication, and node trust. Within this model, the Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) serves as a predictor of potential hidden state sequences, while the Long Short-Term Memory (LSTM) elucidates the relationship between these states and trust levels. Moreover, GHL-SAR deploys an adaptive routing mechanism rooted in the Particle Swarm Optimization Algorithm (PSOA), judiciously weighing link quality for routing decisions. The protocol further advances a density-based spatial clustering method for effective trust evidence aggregation. The simulation results demonstrate that GHL-SAR significantly reduces packet loss and energy consumption, while ensuring detection accuracy and network security compared to other existing routing protocols.

A series of datasets and models have been proposed for summaries generated for wellformatted documents such as news articles.Dialogue summaries, however, have been under explored.In this paper, we present the first dataset with fine-grained factual error annotations named DIASUMFACT.We define finegrained factual error detection as a sentencelevel multi-label classification problem, and we evaluate two state-of-the-art (SOTA) models on our dataset.Both models yield sub-optimal results, with a macro-averaged F1 score of around 0.25 over 6 error classes.We further propose an unsupervised model ENDERANKER via candidate ranking using pretrained encoder-decoder models.Our model performs on par with the SOTA models while requiring fewer resources.These observations confirm the challenges in detecting factual errors from dialogue summaries, which call for further studies, for which our dataset and results offer a solid foundation. 1Lilly: Wanna go out tonight? Marshall: can't :( money's low Lilly: my treat :) Marshall:

Underwater acoustic sensor networks (UASNs) have emerged as a promising technology for marine exploitation. However, due to the challenging characteristics of the underwater communication, including propagation delay and unstable link quality, UASNs face numerous issues. Furthermore, the open nature of the underwater environment poses significant security threats, making secure routing in UASNs a crucial requirement. In this paper, we propose a novel and secure routing approach, named GHL-SAR, to address these challenges. GHL-SAR models and measures the trust level of nodes by considering energy trust, communication trust, and node trust. Specifically, Gaussian Mixture Model-Hidden Markov Model (GMM-HMM) is employed to predict the most likely hidden state sequence for a given observation sequence, and then Long Short-Term Memory (LSTM) is used to determine the relationship between the hidden state and the trust level. Moreover, GHL-SAR utilizes an adaptive routing method based on the Particle Swarm Optimization Algorithm (PSOA), which takes into account the accurate link quality to determine the routing strategy. The simulation results demonstrate that GHL-SAR significantly reduces packet loss and energy consumption while ensuring network security compared to other existing routing algorithms.