The multiple Uncrewed Aerial Vehicles (multi-UAV) collaborative system significantly augments perception capabilities and coverage range of task environments through the establishment of a comprehensive three-dimensional monitoring network, emerging as an indispensable technological cornerstone for future Mobile Crowdsensing (MCS) systems. However, the co-existence of environmental dynamics and device heterogeneity induces non-trivial energy efficiency imbalances across the UAVs, posing a substantial challenge to achieving sustained and efficient multi-UAV exploration. Therefore, we propose an energy-efficient cluster cooperative exploration method that jointly optimizes information sharing and task allocation. To balance communication energy efficiency among UAVs, we introduce an energy-efficient hierarchical information sharing mechanism that dynamically adjusts relay nodes based on real-time attributes of UAVs. In order to improve the utilization of resources, a multi-UAV cooperative task allocation model was developed using cooperative game. It has also been proven that a fair task allocation strategy exists, which is acceptable to all UAVs. Furthermore, the approximate Shapley value of every UAV is calculated using the improved Monte Carlo sampling method combined with incremental update mechanism to ensure fair task allocation. Experimental results demonstrate that the maximum enhancement of task completion ratio is 12%, 26%, and 10%, respectively, at task-critical thresholds for systems utilizing 5, 10, and 15 UAVs. Moreover, the proposed method demonstrated superior performance in energy consumption ratio, synergy, and energy consumption difference compared to benchmarks.

Detecting deviant language such as sexism, or nuanced language such as metaphors or sarcasm, is crucial for enhancing the safety, clarity, and interpretation of social interactions. While existing classifiers deliver strong results on these tasks, they often come with significant computational cost and high data demands. In this work, we propose Class Distillation (ClaD), a novel training paradigm that targets the core challenge: distilling a small, well-defined target class from a highly diverse and heterogeneous background. ClaD integrates two key innovations: (i) a loss function informed by the structural properties of class distributions, based on Mahalanobis distance, and (ii) an interpretable decision algorithm optimized for class separation. Across three benchmark detection tasks – sexism, metaphor, and sarcasm – ClaD outperforms competitive baselines, and even with smaller language models and orders of magnitude fewer parameters, achieves performance comparable to several large language models. These results demonstrate ClaD as an efficient tool for pragmatic language understanding tasks that require gleaning a small target class from a larger heterogeneous background.

Verifying health claims amid rampant misinformation requires identifying qualified experts - a manual process that cannot scale. We address this challenge with a computational framework that automatically identifies biomedical researchers to evaluate health claims by analyzing their PubMed publication profiles. We establish the first benchmark for this cross-genre retrieval task, linking 93,404 health claims to 153,147 biomedical experts. Our two-stage neural pipeline addresses the semantic heterogeneity between informal health claims and formal research literature. Systematic evaluation reveals a striking finding: domain-specific models achieve 84.2% Mean Reciprocal Rank, substantially outperforming general-purpose alternatives, including state-of-the-art LLM-based rerankers fine-tuned on domain-specific data. Our findings underscore the necessity of specialized benchmarks for cross-genre information retrieval, and specialized pretraining for biomedical expert identification, while our scalable architecture enables rapid, automated expert matching for evidence-based claim verification in clinical and public health contexts.

In industrial cyber–physical systems (ICPS), real-time condition monitoring of wear-prone components of steel rolling production equipment is a key scenario for predictive maintenance. Machine vision-based crack detection can quickly identify critical damage and prevent unplanned downtime. However, the harsh working environment poses difficulties for data collection, a large amount of noise tends to contaminate surface crack images, and complex surface crack morphology affects the recognition accuracy. The real-time and accuracy performance of traditional crack detection algorithms are hard to meet the requirement of industrial applications. To tackle this challenge, a high-precision surface crack detection architecture for rolling steel production equipment based on image semantic segmentation is proposed. First, a coordinate attention-deep convolution generative adversarial networks (CA-DCGANs)-based data augmentation method is proposed to augment the original data set with high quality. Second, a crack detection model based on multiscale learning efficient spatial pyramid network (MLESPNetV2) is proposed. It effectively improves detection accuracy to obtain semantic information strongly correlated with crack using multiscale modeling and attention mechanism. Third, A semi-supervised learning method based on multiscale learning efficient spatial pyramid-generative adversarial network (MLESP-GAN) is proposed to solve the problem of insufficient labeled data and unstable training process. Finally, extensive experimental results on KolektorSDD and CAS-Crack data sets demonstrate that the proposed MLESPNetV2 significantly improves accuracy and real-time performance compared with the benchmark model. It is therefore suitable for deployment in industrial sites for real-time health monitoring of industrial equipment.

In intelligent transportation systems, efficient traffic management and population monitoring is attributed to the real-time scheduling of multiple Transportation Big Data (TBD) flows, which strongly supported risk assessment and control during the COVID-19 pandemic. However, as a complex and heterogeneous network, it is difficult to meet the priority and real-time requirements of multi-modal TBD flow scheduling. To improve the real-time performance of TBD flow scheduling, a scheme for Time-Triggered (TT) flow scheduling and Audio-Video Bridging (AVB) flow scheduling is proposed. First, an online scheduling method for TT flows (RFSD) is proposed, which uses Lion Swarm Optimization (LSO) algorithm for priority assignment and dynamic queues to adjust the scheduling order in real time. It ensures fairness in scheduling and effectively improves the utilization of time slot resources. Furthermore, an online scheduling method for AVB flows (RFSU) is proposed, which uses the Imperialist Competitive Algorithm (ICA) to construct the utility function for evaluating the scheduling value of AVB flows, effectively increasing the throughput of AVB flows. Finally, extensive experiments show that RFSD increases successful scheduling by 22% over the PAS algorithm. Compared to the TTA algorithm, RFSU achieves a 24% reduction in average delay and a 27% reduction in jitter.

The Maritime Internet of Things (MIoTs) provides intelligent information services for marine scientific research, emergency response and environmental monitoring by leveraging its wide coverage and ubiquitous connectivity. However, challenging maritime communication conditions and limited sea-based network resources hinder MIoT from meeting the evolving network quality of service requirements of growing maritime activities. This poses a significant challenge to ensuring real-time and reliable transmission of mixed traffic flows. To address issues such as link contention and transmission delays in software-defined MIoT systems, a deterministic scheduling and highly reliable routing mechanism based on cross-layer design is proposed. First, a deterministic scheduling mechanism for mixed traffic flows is introduced, which effectively reduces transmission delays and improves the schedulability of data flows. Second, a high-reliability, low-latency routing mechanism based on Double Deep Q Network (DDQN) is proposed, which is capable of dynamically screening neighbouring nodes based on real-time link and node states, thus facilitating fast and high-quality path selection. Extensive simulation results show that DSMTF improves flow schedulability by 28% compared to traditional algorithms, while HRLDQ increases the network packet delivery rate by 25.8% and reduces the average end-to-end delay by 23.6%.

This paper addresses the challenges associated with storing large amounts of state data on a blockchain by proposing a scalable verifiable ledger database. The rise of digital cryptocurrencies has drawn attention to blockchain technology and its potential benefits in terms of data security, system stability, and trust facilitation. However, existing blockchain-oriented systems face limitations in terms of performance and verifiability. Similarly, traditional ledger databases lack the ability to ensure verifiability of state data. To overcome these challenges, the proposed system introduces a scalable verifiable independent architecture named VI-Store, an improved Merkle tree structure called MMB-tree for state data at the billion level. The evaluation results demonstrate the effectiveness of the proposed system, MMB-tree outperforms traditional MBT and MPT index structures, and VI-Store is able to provide stable storage of billion-level state data over a 7*24 period. The findings suggest VI-Store can be used in real blockchain environments.

As an important index of aero-engine, Remaining Useful Life (RUL) is the key content of prediction. Due to the good generation characteristics of Variational Auto-encoder (VAE) and Generation Adversarial Network (GAN) networks, this paper proposes a Health Index (HI) curve generation method based on VAE-GAN. After that, sensor sequence prediction is carried out through Bidirectional Long Short-Term Memory Network (BLSTM). The two networks are parallel, and then RUL prediction is carried out by synthesizing the data of the two networks. As far as the author knows, this is the first use of VAE-GAN in Prognostics Health Management (PHM). It is verified on the Commercial Modular Aero-Propulsion System Simulation (C-MAPSS) dataset. Finally, the results show that the VAE-GAN network is effective and superior in RUL prediction. At the same time, the proposed parallel network is superior to other RUL prediction methods by generating HI curves.

As part of the recent advances in the IoT world, constrained platforms featuring more than one radio interface have emerged. Typically, they combine the more classical short-range radio and the long or medium range technologies. However, their full potential is not yet exploited. The modifications to the IPv6 routing protocol for Low-Power and Lossy Networks (RPL) and the Medium Access Control (MAC) layer presented in this paper are a first step in this direction. They allow the node to automatically select the most suitable radio link if more than one is available, improving network reliability. This can be especially relevant for monitoring and smart cities applications. The solution is implemented in the ContikiOS, and tested with the Zolertia Re-mote platform, but can be extended to any platform featuring more than one radio interface. The presented testbed experiments indicate that more investigation is needed to optimally tune RPL's metrics for networks featuring nodes with dual-band communication capabilities.

The capability of noninvasive and precise micromanipulation of sensitive, living cells is necessary for understanding their underlying biological processes. Optical tweezers (OT) is an effective tool that uses highly focused laser beams for accurate manipulation of cells and dielectric beads at microscale. However, direct exposure of the laser beams on the cells can negatively influence their behavior or even cause a photo-damage. In this paper, we introduce a control and planning approach for automated, indirect manipulation of cells using silica beads arranged into gripper formations. The developed approach employs path planning and feedback control for efficient, collision-free transport of a cell between two specified locations. The planning component of the approach computes a path that explicitly respects the nonholonomic constraints of the gripper formations. The feedback control component ensures stable tracking of the path by manipulating the cell using a set of predefined maneuvers. We demonstrate the effectiveness of the approach by transporting a yeast cell using four different types of gripper formations along collision-free paths on our OT setup. We analyzed the performance of the proposed gripper formations with respect to their maximum transport speeds and the laser intensity experienced by the cell that depends on the laser power used.

Studying collective migration of cells is currently of considerable interest in biology and medicine leading to possibility of novel diagnosis and treatments. Some cells are highly sensitive to direct laser exposure, which may influence their behavior or even cause photodamage. In addition, manual manipulation of cells is time consuming making it hard to carry out systematic studies that are properly timed to exhibit the desired motility. This paper presents an automated planning approach for precise, collision-free, indirect manipulation of cells with irregular, dynamically changing shapes using Optical Tweezers (OT). We have evaluated the effectiveness of our manipulation approach using physical experiments. We have also carried out an experimental study to demonstrate the effect of the indirect manipulation approach on cell-viability.

In this paper, we present a physics-aware, planning approach for automated transport of cells in an optical tweezers-assisted microfluidic chamber. The approach can be used for making a uniform distribution of cells inside the chamber to allow the study of a variety of biological processes, including cell signaling. Fluid forces inside the chamber, modeled using computational fluid dynamics, are incorporated into the widely used Langevin equation to simulate the motion of cells. The developed simulator was used for building a map that contains probabilities of a cell successfully reaching one of the outlets of the chamber from different locations under the influence of the fluid flow. The developed planner not only generates collision-free paths that exploit the fluid flow inside the chamber but also utilizes the offline generated simulation data to decide suitable locations for releasing the cells. This ensures fast and robust cell transport, while minimizing the required laser power and operational time. The planner is based on the heuristic D* Lite algorithm that employs a specific cost function for searching over a novel state-action space representation. The effectiveness of the planning algorithm is demonstrated using both simulation and physical experiments in a microfluidics-optical tweezers hybrid manipulation setup.

Optical Tweezers (OT) are used for highly accurate manipulations of biological cells. However, the direct exposure of cells to focused laser beam may negatively influence their biological functions. In order to overcome this problem, we generate multiple optical traps to grab and move a 3D ensemble of inert particles such as silica microspheres to act as a reconfigurable gripper for a manipulated cell. The relative positions of the microspheres are important in order for the gripper to be robust against external environmental forces and the exposure of high intensity laser on the cell to be minimized. In this paper, we present results of different gripper configurations, experimentally tested using our OT setup, that provide robust gripping as well as minimize laser intensity experienced by the cell. We developed a computational approach that allowed us to perform preliminary modeling and synthesis of the gripper configurations. The gripper synthesis is cast as a multi-objective optimization problem.