With the expansion of marine development and the occurrence of underwater accidents, the strategic value and practical urgency of searching for wreckage, such as shipwrecks and aircraft, and marine emergency rescue missions continue to rise. However, traditional underwater detection methods have inherent limitations in weak signal resolution and positioning accuracy, severely hindering the accurate identification of underwater debris and the efficiency of emergency responses. To address this issue, an underwater target detection system based on singlephoton imaging technology is presented, employing a technical approach that integrates a noise model-driven data augmentation strategy with deep learning network training. By constructing a multidimensional noise coupling model, integrating typical underwater noise characteristics such as photon shot noise and water scattering noise, a highly realistic virtual underwater singlephoton echo dataset is synthesized, providing standardized data support for model training and testing. Furthermore, by deeply integrating the virtual dataset with the target's physical geometric characteristics, such as debris outline and scale parameters, highprecision detection of underwater debris is achieved. Through the detection of wreckage, the test results show that the recognition accuracy of shipwrecks and aircraft wreckage reaches 0.67, indicating that this method can effectively address engineering problems such as the high cost of acquiring underwater target detection data, insufficient sample diversity, and low efficiency of underwater wreckage detection and emergency rescue.

Battery fault diagnosis is a prerequisite for safety certification to electric vertical take-off and landing (eVTOL) aircraft. Unlike ground vehicles, eVTOL systems operate under a zero-failure tolerance regime, as even minor battery anomalies can compromise flight safety. However, existing diagnostic methods exhibit two fundamental limitations: their structural dependence on single-fault assumptions and the limited adaptability to high-rate electro-thermal coupling under aviation duty cycles. To address these challenges, this study proposes a structurally decoupled, model-based diagnostic framework for battery modules. A cell-level electro-thermal coupling model is established to capture dynamic electrical-thermal interactions under high discharge rates. Four minimal structurally over-constrained subsystems are constructed to generate analytically decoupled residuals. This design enables systematic isolation of seven fault categories, including short circuits, interconnection faults, and multiple sensor failures, while preserving diagnosability under concurrent fault conditions. Kalman filtering enhances state estimation robustness, and an adaptive threshold strategy accommodates varying operational regimes. Under single-fault scenarios, the proposed method achieves a detection rate of 93.88%, enabling isolation and parameter estimation of all seven fault types. More critically, under concurrent-fault scenarios in a 10S3P module with 4323 possible fault-pair combinations, the types of individual faults can be identified in 85.5% of cases, and in 98.3% of cases, the types can be narrowed down to fewer than three candidates.

As pivotal mobile nodes within the Marine Internet of Things (MIoT), Unmanned Surface Vessels (USVs) rely on autonomous berthing as a critical service to ensure persistent data offloading and energy replenishment. However, this maneuver demands precise navigation within confined, high-risk environments, where traditional edge control methods often lack robust safety guarantees. To address these challenges, this paper presents a hybrid safety-aware reinforcement learning (RL) framework tailored for autonomous berthing. By directly processing raw LiDAR data, the proposed end-to-end policy eliminates the reliance on a priori global maps, thereby enhancing adaptability in unstructured environments. To overcome the inherent safety limitations of standard RL, we first introduce the Soft-Constrained Policy Optimization (SCPO) paradigm. Incor-porating a safety critic and a Lagrangian dual update mechanism, SCPO guides the agent to internalize safety constraints during the learning process. Furthermore, to bridge the gap between discrete decision-making and the continuous safety assurance required for docking, we propose a Hard-Constrained Safety Filter (HCSF). Grounded in Control Barrier Functions (CBFs), this module enforces safety constraints via real-time Quadratic Programming (QP) corrections on control inputs. Simulation results demonstrate that the proposed framework exhibits superior safety performance compared to existing baselines.

A major challenge in ensuring the reliability of battery systems is the uncertainty surrounding their service life. An accurate prediction of the remaining useful life (RUL) is essential for effective maintenance and operation. Traditional extended Kalman filter (EKF) algorithms, which rely heavily on historical data, often have limited long-term prediction accuracy. To overcome this problem, a multi-observation fusion approach is proposed to enhance the performance of the EKF for battery life prediction. To improve the practical applicability, the conventional aging capacity test values with operational capacity data obtained under real-world conditions are replaced. This modification refined the trajectory of the battery aging state space, thereby reducing the dependency on the quantity and quality of historical data while simultaneously boosting the long-term prediction accuracy and stability. Furthermore, a semi-empirical aging model is introduced to extract prior knowledge from offline data. This provided valuable insights into the life degradation trends and guided the filtering process. The resulting framework forms the basis for a novel RUL prediction method that utilizes a multi-observation fusion EKF. Validation experiments show that the proposed method enhanced the prediction accuracy by over 60% compared with traditional EKF and particle filter algorithms throughout the lifecycle of lithium-ion batteries. Additionally, the technique exhibited robust stability (with no divergence observed over the full battery life) and demonstrated notable improvements in early stage RUL prediction.

To address the characterization biases of traditional radiative transfer equations and their approximate models in short-range underwater single-photon detection scenarios, this paper proposes a multi-dimensional modeling improvement for single-photon lidar observation systems, considering time, space, and phase. In the time dimension, cumulative absorption modeling based on higher-order scattering path lengths is introduced to correct the survival probability and energy proportion of scattered photons at each order. In the spatial dimension, a particle radial distribution function and a structure factor are introduced to correct the angular distribution of the scattering phase function. In the phase dimension, a mechanism for regulating absorption on optical coherence and weak localization effects is established, enabling the separate expression of coherent and incoherent transmission channels. Based on these, a parameterized model for single-photon detection is constructed, and calculable constraints for time-gated width, adaptive sampling step size, and travel error compensation are derived, achieving a mapping from optical transmission physical quantities to system parameters. An underwater single-photon lidar system is built based on this model, and experimental verification is conducted. The results show that the established model provides a data foundation for system timing control, scanning design, and point cloud reconstruction and exhibits stable performance improvements in actual underwater imaging.

The search for sunken ships, aircraft wreckage, and other underwater debris, as well as conducting marine emergency rescue operations, are becoming increasingly significant. However, traditional detection methods suffer from limitations such as weak signal detection capability and low detection accuracy, which severely hinder the precise detection of underwater debris and the efficiency of emergency rescue operations. To address this issue, a biologically inspired hierarchical trapping mechanism is presented. By mimicking the trapping behavior of deep-sea anglerfish using bioluminescent lures, a “global environment perception - local edge enhancement - dynamic strategy optimization” architecture is designed to improve the network's feature extraction capability. This is then combined with underwater single-photon LiDAR to achieve precise detection of underwater debris. The test results show that the presented method achieves an accuracy of 86.2% in identifying sunken ships and aircraft wreckage, representing a 29% improvement compared to traditional models trained with real-world data. This shows that the proposed method effectively addresses engineering challenges, including the high cost of acquiring underwater target-detection data, insufficient sample diversity, and low efficiency in underwater debris detection and emergency rescue operations.

Object detection based on multimodal images is an emerging method in remote sensing which has gained extensive attention in recent years, particularly for all-weather robust detection applications. However, in multimodal remote sensing object detection, there are issues of overreliance on a single modality and intermodal interference. To solve these problems, this article proposes a complementary mask-enhanced transformer detection network (CMTDet). Two key modules are designed in CMTDet: the dual-stage cross-modal interaction (DCMI) module and the common-mode guided weighted fusion (CMGF) module. The DCMI module consists of two components: the differential guided complementary enhancement (DGCE) submodule and the complementary mask cross-attention (CMCA) mechanism. The DGCE component captures differential information from features of different modalities, enhancing the ability to express the semantic-channel correlation and differential information between modalities. Then, the CMCA component explores the correspondence between features of different modalities through cross-attention and reduces the dependence of feature extraction on a single modality using asymmetric complementary masks. After the improved feature extraction, the CMGF module generates guidance signals based on common-mode information, explores common information between modalities and the contribution of information across different modalities, and designs adaptive weights to retain the advantageous information of each modality. Ultimately, this enhances the semantic expression ability and robustness of the fused modality. Extensive experiments on the DroneVehicle, VEDAI and OGSOD datasets show that CMTDet outperforms the baseline: it achieves 1.87% and 1.73% higher mAP50 and mAP50-95 on the DroneVehicle dataset, with a 1.98% mAP boost in nighttime scenarios, verifying its superior detection performance.

In safety-critical systems, single-event upsets (SEUs) in stored weights can corrupt convolutional neural network (CNN) inference. Full triple modular redundancy (TMR) masks single-bit faults but adds 200 -0.03± 0.10 percentage points on LeNet-5 and -1.54± 0.41 percentage points on VGG16. On LeNet-5, Selective and Full DABP mitigate 72.65

Battery fault diagnosis is a prerequisite for safety certification in electric vertical take-off and landing (eVTOL) aircraft. Unlike ground vehicles, eVTOL systems operate under a zero-failure tolerance regime, where even minor battery anomalies can compromise flight safety. However, existing diagnostic methods exhibit two fundamental limitations: their structural dependence on single-fault assumptions and limited adaptability to high-rate electro-thermal coupling under aviation duty cycles. To address these challenges, this study proposes a structurally decoupled, model-based diagnostic framework for battery modules. A cell-level electro-thermal coupling model is established to capture dynamic electrical–thermal interactions under high discharge rates. Based on structural analysis, four minimal structurally over-constrained subsystems are constructed to generate analytically decoupled residuals. This design enables systematic isolation of seven fault categories, including short circuits, interconnection faults, and multiple sensor failures, while preserving diagnosability under concurrent fault conditions. Kalman filtering enhances state estimation robustness, and an adaptive threshold strategy accommodates varying operational regimes. Under single-fault scenarios, the proposed method achieves a detection rate of 93.88%, enabling complete isolation and parameter estimation of all seven fault types. More critically, under concurrent-fault scenarios in a 10S3P module with 4,323 possible fault-pair combinations, 85.5% are uniquely identified and 98.3% are reduced to fewer than three candidates. These results demonstrate that structural residual design, rather than data-driven pattern recognition, is essential for achieving certifiable multi-fault diagnosability in next-generation eVTOL battery systems.

Aiming at the shortcomings of traditional magnetic positioning technology in feature extraction ability and positioning efficiency, this paper proposes a road positioning method based on deep learning fusion of magnetic information and inertial unit. By fusing geomagnetic data and inertial information, this method eliminates redundant data and realizes multi-dimensional magnetic signal analysis, which significantly improves the feature discrimination of different positions. Aiming at the problem of high construction cost of the geomagnetic fingerprint database, a geomagnetic sequence enhancement algorithm is designed, which can simulate the generation of data sets samples collected by different driving speeds and devices, and effectively improve the adaptability of the model in complex scenes. In addition, in order to solve the problem of insufficient processing ability of traditional neural networks for spatio-temporal features, the DM-CHLSTM model is proposed to accurately capture spatio-temporal dependencies in long-distance positioning through the fusion of convolution and long short-term memory network. Two smartphones IQOO Neo8 and OPPO Find X8 with different built-in magnetic sensors were used for multiple rounds of verification on three paths covering different scenarios. The results show that compared with the mainstream deep learning methods, the average positioning error of the proposed algorithm is reduced by 78.01% and RMSE is reduced by 79.38% at most, and it shows good robustness between different devices, which provides an efficient solution for high-precision road positioning.

The single-event effect has a great influence on the performance of the electronic devices. In order to improve the reliability of the track detectors applied in radiation environments, two radiation-hardened latch designs based on magnetic tunnel junction (MTJ) are proposed in this article. The proposed latches employ the hybrid CMOS/MTJ circuit structure to achieve nonvolatility. To address the issue of insufficient radiation tolerance in traditional precharge sense amplifier (PCSA) circuits, an improved PCSA architecture (IPCSA) is adopted. The proposed PLATCH-1 employs two copies of the IPCSA units and a couple of two-input C-elements to achieve single-node-upset (SNU) recovery, while PLATCH-2 employs three identical copies and two three-input C-elements to achieve double-node-upset (DNU) tolerance. Simulation results show that when the radiation charge is 2 pC, the PLATCH-1 latch recovers from all the SNUs and the PLATCH-2 latch is all DNUs tolerant. This work provides a feasible design for the utilization of MTJ memory devices in the sequential logic circuits and holds great significance in the high energy physics experiment applications within aerospace radiation environments.

To solve the fast fixed-time consensus tracking problem for high-order multiagent systems, this paper proposes a control method to enable efficient consensus among multiple agents. A novel distributed high-order fixed-time observer is designed for each follower to accurately estimate the state of a leader. A fixed-time stable system with a faster convergence rate than existing systems is developed. Based on this system and the observer, a novel fast fixed-time consensus tracking control strategy is proposed. The suggested control protocol guarantees that (1) the estimated leader's state is accurately tracked by the followers and the convergence rate of the high-order system errors is expedited; (2) the system tracking errors can be maintained near the origin in a fixed time, and importantly, the upper bound on the convergence time of the system tracking errors is estimated regardless of the initial state. Simulation results indicate the effectiveness and superiority of the proposed approach.

Underwater single-photon LiDAR (SP-LiDAR) systems are strongly influenced by photon attenuation and backscattering in complex aquatic environments, which can reduce ranging accuracy and imaging resolution. This effect is mainly caused by strong absorption and scattering from water molecules and suspended particles, especially in turbid waters. To address this issue, we propose a signal reconstruction algorithm based on multi-variational waveform decomposition and guidance (MVWDG), where the guidance refers to the incorporation of physical characteristics of underwater photon propagation, backscattering behavior, and photon-counting noise statistics into the signal processing framework. The algorithm utilizes these physical priors to efficiently extract target information and suppress noise, allowing the reconstruction of target depth and reflectivity-related intensity (RRI) from complex photon echoes. The method consists of three stages: data preprocessing, variational mode decomposition (VMD), and collaborative Optimization, achieving high-quality target reconstruction in challenging underwater conditions. Experimental results from our SP-LiDAR system show that the MVWDG algorithm outperforms existing imaging methods, achieving a lateral resolution of approximately 5 cm and a range resolution of 2.5 cm@6AL, demonstrating its advantages for high-resolution underwater imaging and indicating its potential as a reliable technique for underwater target detection.

Reconstructing subsurface sea temperature and salinity from multi-source surface observations is important for understanding ocean dynamics and environmental variability, but it remains challenging because of strong spatial heterogeneity and complex spatiotemporal coupling. This paper presents MSFFNet, a spatiotemporal reconstruction network for subsurface thermohaline estimation. The proposed model consists of three coordinated components. A spatial feature extraction branch combines depthwise separable convolution with coordinate attention to enhance the representation of local gradients and direction-aware spatial structures. A temporal feature extraction branch adopts multiscale temporal modeling to capture historical variations at different temporal resolutions while preserving sequence-level temporal information. A spatiotemporal fusion block then performs bidirectional interaction between spatial and temporal features, followed by temporal conditioning and attention-based refinement, so that the two types of information can be integrated more effectively before final prediction. Experiments on western Pacific data from 2000 to 2020 demonstrate that MSFFNet consistently outperforms several competing baselines in both temperature-anomaly and salinity-anomaly reconstruction. The results further show that the proposed spatial enhancement and spatiotemporal fusion strategies make substantial contributions to the overall performance. These findings confirm the effectiveness of MSFFNet for subsurface thermohaline reconstruction in complex ocean environments.

In radiation environment, the single event effect (SEE) poses great threat to the peripheral circuit of the particle detectors. In order to improve the anti-radiation performance of the CMOS pixel sensor, a double-node-upset-hardened full-subtractor applying magnetic tunnel junction (MTJ) is proposed in this paper. To improve the radiation resistance of the peripheral circuit, a redundancy transistor based peripheral circuitry structure is designed. The working state of the circuit is controlled by the designed reading-writing control circuit. Different from the conventional writing circuit, it controls the writing sequence by an OR gate. This circuit could control the reading and writing to switch flexibly and reduce the writing supply voltage. Simulation results show that the proposed circuit functions as a full-subtractor and the maximum double-node-upset tolerant charge is over 1000fC. This work presents a feasible candidate for the application of MTJ in the digital circuit of CMOS pixel sensors and has great prospect for the high energy physics experiments.

Provides society information that may include news, reviews or technical notes that should be of interest to practitioners and researchers.

Single-event upsets (SEUs) significantly threaten the reliability of convolutional neural network (CNN) accelerators in aerospace systems by corrupting the floating-point weights stored in memory. This paper systematically analyzes the bit-level sensitivity of IEEE 754 single-precision CNN weights through rigorous theoretical modeling and extensive fault injection experiments covering more than 150,000 trials. It proposes selective memory protection methods based on the identified sensitivity patterns. Our key finding reveals a previously unreported non-monotonic bit-sensitivity phenomenon, wherein certain middle exponent bits (b26, b25, b24) exhibit higher error vulnerability compared to traditionally prioritized higher-order bits. This insight enables two innovative memory protection schemes targeting only the five most sensitive bits (b31, b30, b26, b25, b24), reducing memory overhead from 200% to 31.25%, or even eliminating extra memory usage by embedding redundancy within low-sensitivity bits. These findings pave the way for developing highly reliable yet resource-efficient CNN accelerators tailored for severe radiation environments.

Addressing the demand for long-range, highresolution target reconstruction of underwater objects, this study investigates single-photon lidar target imaging technology, presenting a multi-stage algorithmic framework for single-photon lidar imaging. This framework possesses strong information extraction and noise filtering capabilities, enabling the reconstruction of target depth and reflectivity information from complex echo photons. The method comprises three steps: data preprocessing, variational modal decomposition, and collaborative variational imaging enhancement, achieving highquality target reconstruction in complex underwater environments. Test results demonstrate that this target reconstruction method, utilizing the constructed single-photon lidar system, achieves a centimeter-level lateral and range resolutions at 5-6 AL, indicating significant advantages for highresolution underwater imaging and providing technical support for underwater target detection.

Bolts are widely applied in critical sectors such as industrial manufacturing, construction, and transportation, where they play a key role in large-scale mechanical equipment. The fastening condition of bolts directly affects the stability and safety of equipment operation. To address the challenge of timely detecting bolt loosening, this paper proposes a real-time bolt loosening detection method based on an improved YOLOv8 algorithm combined with machine vision technology. Through comparisons with other object detection algorithms, including Faster R-CNN, EfficientDet, SSD, and DETR, YOLOv8 was selected as the baseline model due to its superior performance in balancing detection accuracy and computational efficiency. The YOLOv8 model was further optimized through lightweight redesign and pruning, reducing the parameter count to 13

This research investigates the application of Integrated Sensing and Communication (ISAC) systems in maritime secure communication networks. Considering the severe path loss caused by complex maritime environments, we employ an active Reconfigurable Intelligent Surface (RIS) to establish supplementary communication links and enhance system performance. The primary objective is to maximize the sum secrecy rate (SSR) through the joint optimization of the hybrid precoding at the Dual-Function Base Station (DFBS) and the phase shift matrix of the active RIS, subject to both transmit power constraints and sensing performance requirements. We propose an alternating optimization algorithm based on fractional programming (FP) and successive convex approximation (SCA) techniques to solve this joint optimization problem. Numerical results demonstrate that the proposed algorithm achieves up to 10% SSR gain compared to passive RIS schemes, while maintaining excellent convergence performance.