In Unmanned Aerial Vehicle (UAV), aerial photography, hardware and imaging principle differences between multi-sensor systems cause significant variations in image resolution, field of view (FOV), and modality, leading to issues such as uneven feature point distribution, low matching rates, and local deformation in image registration. To address these, we propose a multi-stage infrared (IR) and visible (VIS) image registration pipeline. First, in the preprocessing stage, to mitigate scale inconsistencies between IR and VIS images, we introduce the Phase-Consistent Gaussian Total Variation (PC-GTV) method, which maps the original image into a Phase Consistency (PC) space. This intermediate domain aligns local phase information to unify edge structures and geometric contours. A weighting strategy combining Root Mean Square Error (RMSE) and Structural Similarity Index Measure (SSIM) is then used for scale estimation to determine the optimal scaling between images. Second, in the coarse registration stage, to overcome the uneven distribution of traditional feature points, a chunk-based detection strategy is applied in the Phase Consistent space to enhance texture response and improve feature point uniformity. Finally, in the fine registration stage, to tackle low matching rates and local distortion, a coupled constraint strategy based on the Multiscale Cyclic Oriented Gradient (MCOG) feature map is used to estimate non-rigid offsets, while an elastic compensation model completes the fine registration. Extensive experiments demonstrate that our method achieves strong visual and quantitative performance on public datasets. Our code is available at https://github.com/lh-ite/UAV_IRVIS_Registration.

We propose an U-IoT sensing network based on Fiber-Optic DAS, with SSWF enhancing signals and array network enabling near-field localization. Experiments show 6.3 dB SNR improvement, as well as MVDR localization achieves 0.08 m and 5.04° errors. © 2026 The Author(s)

Evolutionary multitasking optimization (EMTO) has emerged as a powerful tool for addressing complex optimization problems, particularly those involving multiple related tasks. However, current EMTO methods prioritize promoting convergence characteristics while paying little attention to potential privacy leakage issues during knowledge transfer among tasks. This paper designs privacy-preserving centralized and decentralized EMTO models based on inter-task knowledge transfer, and proposes a perturbation query strategy for the centralized EMTO. Furthermore, passive and active attacks in EMTO are proposed for the first time, and an example of active attacks in EMTO is provided by constructing a deceptive malicious task. The active attack technique can serve as a test to determine whether the EMTO methods can effectively identify and defend against malicious knowledge transfer. Experimental results show that the proposed privacy-preserving centralized EMTO method can protect clients’ privacy information while maintaining algorithm performance, and the proposed active attack technique can significantly slow down the convergence speed of the target task.

This paper introduces an Elliptically-Wrapped Vector Sensor (EWVS) that breaks the scalar limitation of conventional DAS by demonstrating a distinct vector response. Experimental results confirm its directional discrimination capability while maintaining high signal fidelity.

Deep neural networks are vulnerable to adversarial attacks in infrared object detection. Previous infrared physical attacks have primarily targeted 2D pedestrian detection, showing limited practical effectiveness. Therefore, this paper proposes a design named Adversarial Electrothermal Clothes (AdvETC) based on a 3D human model, to attack infrared detectors from multiple viewpoints. This paper introduces a deep reinforcement learning-based optimization framework for AdvETC, enabling pedestrians to evade detection from various angles and distances. First, a 3D human model is created based on the infrared characteristics of humans in the physical world. Next, an artificial potential field is established within the search area, and a reward mechanism is introduced based on the detector's feedback and changes in the field's potential energy. Finally, trajectory points are smoothly connected using tangent-continuous Bézier curves to generate physically feasible and efficient adversarial trajectories. This paper is the first to apply graphene material in infrared adversarial attacks. By incorporating graphene-based heating fabric into the clothing based on the designed adversarial samples, the heating fabric heats up when powered, altering the infrared characteristics of the person’s surface. Due to the physical carrier's stealth, as well as the stability and continuous controllability of infrared radiation, our method significantly enhances attack effectiveness and reduces the complexity of practical implementation. Extensive experimental results show that, compared to other infrared physical attack methods, the AdvETC exhibit superior adversarial performance and robustness.

Untargeted adversarial abundance attacks in hyperspectral image (HSI) classification have received growing attention due to their budget efficiency and physical plausibility, which arises from adherence to spectral mixing principles. However, it is difficult to design an abundance-domain perturbation for targeted attacks to simultaneously suppress evidence for the source class while amplifying evidence for a specified target. This article proposes a targeted adversarial abundance attack (TA(3)) framework that steers reconstructed spectra toward the decision region of the designated class. The proposed method first applies deep learning-based unmixing to obtain endmembers and abundances without reliance on external spectral libraries. A class-guided perturbation strategy is proposed to enforce abundance-vector similarity and feature-space alignment. Then, a gradient conflict resolution mechanism is proposed to balance multiple loss objectives, which stabilizes updates and maintains small, physically consistent perturbations. Experiments on benchmark HSI datasets demonstrate that the proposed targeted adversarial abundance-attack framework achieves targeted success with improved imperceptibility and small perturbation norms.

Machine learning (ML), driven by prominent paradigms such as centralized and federated learning, has made significant progress in various critical applications ranging from autonomous driving to face recognition. However, its remarkable success has been accompanied by various attacks. Recently, the model hijacking attack has shown that ML models can be hijacked to execute tasks different from their original tasks, which increases both accountability and parasitic computational risks. Nevertheless, thus far, this attack has only focused on centralized learning. In this work, we broaden the scope of this attack to the federated learning domain, where multiple clients collaboratively train a global model without sharing their data. Specifically, we present HijackFL, the first-of-its-kind hijacking attack against the global model in federated learning. The adversary aims to force the global model to perform a different task (called hijacking task) from its original task without the server or benign client noticing. To accomplish this, unlike existing methods that use data poisoning to modify the target model's parameters, HijackFL searches for pixel-level perturbations based on their local model (without modifications) to align hijacking samples with the original ones in the feature space. When performing the hijacking task, the adversary applies these cloaks to the hijacking samples, compelling the global model to identify them as original samples and predict them accordingly. We conduct extensive experiments on four benchmark datasets and three popular models. Empirical results demonstrate that its attack performance outperforms baselines. We further investigate the factors that affect its performance and discuss possible defenses to mitigate its impact.

We demonstrate a fiber-optic planar array with a reconfigurable layout for ultrasound detection. The four-sensor system is proposed to verify the performance with an average NEP of 39.1 Pa and fourfold enhancement in imaging efficiency.

Deep neural networks have substantially improved the performance of hyperspectral image classification, yet they remain vulnerable to adversarial attacks. Existing attack methods usually manipulate pixel spectra directly, ignoring the physical mixing mechanism of remote sensing imaging and potentially generating adversarial samples with limited physical consistency and interpretability. Moreover, balancing attack effectiveness and perturbation imperceptibility remains a challenging multi-objective optimization problem. To address these issues, this paper proposes an evolutionary multi-task multi-objective adversarial attack framework based on inter-task knowledge transfer. Instead of perturbing raw pixel spectra, the proposed method introduces perturbations into abundance maps obtained through spectral unmixing, thereby improving the physical plausibility of the generated adversarial samples. The generation of class-specific universal perturbations is formulated as a collaborative multi-task optimization problem. To solve this problem, we develop a Self-Adaptive Multi-Objective Multi-Factorial Evolutionary Algorithm for Adversarial Attacks (SAMO-MFEA-AA). By modeling the attack generation processes for different land-cover classes as distinct yet correlated optimization tasks, SAMO-MFEA-AA dynamically captures synergistic relationships among tasks. An asymmetric adaptive cooperation matrix is employed to regulate the intensity of knowledge transfer, allowing beneficial perturbation patterns to be shared across related classes while reducing the risk of negative transfer. Extensive experiments on the Indian Pines and Salinas datasets demonstrate that the proposed framework achieves competitive hypervolume performance and favorable solution diversity compared with existing multi-objective optimization algorithms. In adversarial attack scenarios, the proposed method achieves effective attack success rates against representative classification networks while maintaining the physical plausibility of abundance-space perturbations.

For expensive multiobjective optimization problems, there exists useful knowledge, e.g., the trained surrogate models, can be transferred to assist the optimization of a target optimization problem, which is termed as multiproblem surrogates. Stacking transfer is able to combine the pretrained source surrogate models and the preliminary target model with a meta-regression algorithm to transfer knowledge from source to target. However, when large-scale source models are involved in the many-problem scenarios, the less correlated sources may hurt the target performance, which is known as negative transfer. In this article, sparse representation of the coefficients of meta-regression is considered to automatically select the most relevant source models for largely avoiding negative transfer. In the proposed many-problem surrogates, the coefficients of the source and target models are assumed to be sparse under the non-negativity and sum-to-one constraints. Then, a sparse transfer stacking model is established with l(1) -norm of the coefficients. Next, the alternating direction method of multipliers is employed to solve the resulting constrained optimization problem by converting it into several much simpler problems. Most of the previous works assume that the costs for evaluation have no much difference and this assumption rarely holds in the real-world applications. In order to further reduce the total costs, an improved surrogate model with a cost-sensitive measure is designed to estimate the cost and select new solutions for real evaluation based on their estimated fitness, uncertainty and cost. Experimental results on synthetic and practical problems have demonstrated the superiority of the proposed many-problem surrogates.

Fiber-optic vibration sensors are crucial for detecting subtle disturbances in applications such as ocean seismic monitoring and structural health assessment. Among various configurations, interferometric fiber-optic sensors offer the advantages of long-range measurement, high dynamic range, and seamless integration into existing telecommunication networks. However, their performance is significantly constrained by laser frequency noise, particularly low-frequency drift, which degrades both sensitivity and signal-to-noise ratio (SNR). To address this issue, a Sampling Matched Integral Fitting Difference (SMIFD) scheme is proposed for low-frequency drift compensation in interferometric fiber-optic vibration sensing systems, leveraging intrinsic system properties and hardware component-free. By synchronizing the sampling rate with the fiber's transmission delay, the randomly distributed low-frequency noise is transformed into phase drift with discernible time-frequency characteristics through integration. Subsequently, the laser frequency drift can be predicted and mitigated from the phase drift through polynomial fitting. The experimental demonstration shows that SMIFD achieves significant noise floor suppression of 81.49 dB at 10 mHz and a SNR enhancement of 27.02 dB at 200 mHz over a 50 km single-mode fiber link. Furthermore, the down-sampling strategy is used to overcome the sampling rate limitation imposed by the transmission delay, enabling a broadband detection from millihertz to kilohertz range. Experimental results confirm that our method achieves the measurement accuracy required for ocean monitoring applications, even in high-drift conditions, enabling precise detection of low-frequency activities.

Cardiorespiratory rhythm disorders (CRD) are prevalent yet underdiagnosed conditions with substantial health risks. Current diagnostic methods are limited by invasiveness, intermittent symptom presentation, and low sensitivity during wakefulness. Sleep, by contrast, offers a stable physiological state ideal for early CRD detection. Here, we introduce a fiber-optic vital sign monitoring pillow (VSMP) that aims to realize the vision of home-based nocturnal health management, leveraging natural sleep for the accurate and unobtrusive early detection of CRD. By integrating fiber-optic sensing, biomechanical engineering, and advanced signal processing, the VSMP achieves early CRD detection through heart rate variability analysis and Lorenz scatter plots, while also facilitating blood pressure estimation (<8% error). Thus, we present VSMP as a new paradigm that transforms passive sleep into an active diagnostic session, paving the way for data-driven, personalized health management and proactive, precision medicine directly in the home environment.

We demonstrate the first long-haul co-wavelength ISAC over a 190 km WDM fiber link, simultaneous achieving over 20 Tb/s transmission capacity and 0.01 Hz ultra-low-frequency sensing with 20 m spatial resolution for transoceanic monitoring.

Evolutionary multiobjective optimization (EMO) has made significant strides over the past two decades. However, as problem scales and complexities increase, traditional EMO algorithms face substantial performance limitations due to insufficient parallelism and scalability. While most work has focused on algorithm design to address these challenges, little attention has been given to hardware acceleration, thereby leaving a clear gap between EMO algorithms and advanced computing devices, such as GPUs. To bridge the gap, we propose to parallelize EMO algorithms on GPUs via the tensorization methodology. By employing tensorization, the data structures and operations of EMO algorithms are transformed into concise tensor representations, which seamlessly enables automatic utilization of GPU computing. We demonstrate the effectiveness of our approach by applying it to three representative EMO algorithms: NSGA-III, MOEA/D, and HypE. To comprehensively assess our methodology, we introduce a multiobjective robot control benchmark using a GPU-accelerated physics engine. Our experiments show that the tensorized EMO algorithms achieve speedups of up to 1113x compared to their CPU-based counterparts, while maintaining solution quality and effectively scaling population sizes to hundreds of thousands. Furthermore, the tensorized EMO algorithms efficiently tackle complex multiobjective robot control tasks, producing high-quality solutions with diverse behaviors. Source codes are available at https://github.com/EMI-Group/evomo.

Photoacoustic (PA) and ultrasound (US) dual-modality imaging, combining the high resolution of PA imaging and the high penetration depth of US imaging, has long been envisioned as an edge tool. This complementary duality holds transformative potential for applications ranging from lymph node characterization to intravascular diagnostics. Yet, conventional dual-modality imaging systems remain constrained by the limited integration complexity. This study pioneers a paradigm-shifting imaging system with a highly integrated single-fiber-based PA-US dual-modality transducer. The core of the system lies in a nontoxic ultraviolet glue-dye (UV-dye) transducer design, which exhibits wavelength-selective optical properties, enabling efficient US generation and PA excitation simultaneously. This transducer architecture achieves an ultrasound generation efficiency of 0.04 MPa mJ-1 cm2 with a 37.1 MHz bandwidth excited by a 532 nm laser pulse, which are 67% and 106% higher than the existing transducer, respectively. Experimental validation is conducted on the established integrated PA-US dual-modality system, in which the lateral resolutions of US and PA imaging modalities are calibrated to be as high as 90 and 125 mu m, while axial resolutions are, respectively, verified as 60 and 55 mu m at a depth of 2.5 mm, effectively equipped as an intelligent microscope for biological tissue imaging. Further, the system successfully reconstructs dual-modal images of ex vivo tissues that transcend traditional single-mode limitations, revealing detailed structural and chromophore information on the biological tissue. The proposed all-fiber dual-modality imaging system owns the capability of high resolution and sufficient tissue details, demonstrating its broad application prospect in angiography and oncology.

This paper presents a contactless cardiorespiratory monitoring System that delivers high-precision acquisition of heart rate (error <2.6 bpm) and respiration rate (error <1.6 bpm), enabling assessment of sympathetic-parasympathetic activity via heart rate variability analysis.

CD44, a well-established cancer biomarker, plays a critical role in tumor invasion and metastasis across multiple cancer types. However, the clinical utility of current CD44 detection platforms is hampered by the requirement for labeling and limited sensitivity, rendering trace-level detection complex and challenging, and thus necessitating the development of novel, label-free, and highly sensitive CD44 sensors. Herein, we propose a silanization-modified label-free fiber biosensor with enhanced sensitivity by tuning the dispersion turning point and high selectivity CD44 recognition via targeted antibody immobilization. A differential tapering technique is adopted to fabricate the micro/nano-fibers with low propagation loss and high spectral contrast, which effectively addresses the drawbacks of traditional fabrication methods such as severe airflow interference and fragile microstructure. Furthermore, a robust covalent immobilization interface for antibodies is constructed via silanization treatment, endowing the sensor with reliable specific binding capability toward CD44. Ultimately, the optimized sensor operates within a dynamic range of 100 fM–6.05 nM, where it exhibits a linear wavelength shift of 3.6443 nm per decade increase in CD44 concentration and achieves a remarkably low detection limit of 0.19 pM. This represents a 1000-fold enhancement in sensitivity compared to existing CD44 detection methods. This highly sensitive and robust CD44 sensor holds significant potential for applications in diagnosis, staging, and prognosis of multiple cancer types.

The rapid growth of uncrewed aerial vehicle (UAV) has increased the demand for secure and efficient training of Artificial General Intelligence models. The Low Altitude Economy (LAE), including precision agriculture, urban mobility, and logistics, further emphasizes the need for efficient machine learning frameworks due to the large volumes of heterogeneous data generated by UAV and constraints in communication bandwidth and energy. In response, Federated Learning (FL) has emerged as a promising solution, offering privacy-preserving capabilities while enabling collaborative model training across distributed devices. This paper introduces an efficient Asynchronous Hierarchical Federated Learning (AHFL) framework, called AchiFL, which supports asynchronous model uploads by dynamically adjusting the clock and clusters clients with similar data distributions for hierarchical aggregation. Although the existing AHFL architecture mitigates prolonged server wait times through asynchronous clocks and alleviates communication bottlenecks via hierarchical aggregation, it remains susceptible to performance degradation due to convergence delays and data heterogeneity at the edge layers. To tackle these challenges, AchiFL incorporates a two-phase optimization approach. First, a clustering-based gradient pruning mechanism groups clients with similar data distributions under edge servers and resolves gradient conflicts using directional pruning and amplification, thereby accelerating convergence under non-Independent and Identically Distributed (non-IID) conditions. Furthermore, through rigorous proof, we identify the key metrics influencing model performance and convergence. By integrating model obsolescence, prediction accuracy, and client participation, we quantify the obsolescence effect at the edge servers and establish a dynamic edge weight penalty mechanism. This mechanism enables adaptive re-weighting of the edge models during the global aggregation process. Extensive experiments show that AchiFL achieves an accuracy improvement of 3.10% to 13.23% on non-IID datasets.