Recently segment anything model (SAM) has attracted widespread concerns, and it is often treated as a vision foundation model for universal segmentation. Some researchers have attempted to directly apply the foundation model to the RGB-D video salient object detection (RGB-D VSOD) task, which often encounters three challenges, including the dependence on manual prompts, the high memory consumption of sequential adapters, and the computational burden of memory attention. To address the limitations, we propose a novel method, namely Segment Anything Model with Depth-guided Adaptive Queries (SAM-DAQ), which adapts SAM2 to pop-out salient objects from videos by seamlessly integrating depth and temporal cues within a unified framework. Firstly, we deploy a parallel adapter-based multi-modal image encoder (PAMIE), which incorporates several depth-guided parallel adapters (DPAs) in a skip-connection way. Remarkably, we fine-tune the frozen SAM encoder under prompt-free conditions, where the DPA utilizes depth cues to facilitate the fusion of multi-modal features. Secondly, we deploy a query-driven temporal memory (QTM) module, which unifies the memory bank and prompt embeddings into a learnable pipeline. Concretely, by leveraging both frame-level queries and video-level queries simultaneously, the QTM module can not only selectively extract temporal consistency features but also iteratively update the temporal representations of the queries. Extensive experiments are conducted on three RGB-D VSOD datasets, and the results show that the proposed SAM-DAQ consistently outperforms state-of-the-art methods in terms of all evaluation metrics.

The convergence of edge intelligence and networked medical infrastructures in the Artificial Intelligence of Medical Things (AIoMT) is transforming healthcare toward personalization and predictive intervention. In this paradigm, high-resolution physiological data continuously flow between wearable or implantable devices, edge nodes, and cloud analytics platforms. Such connectivity enables advanced diagnostic modeling and real-time decision support. However, it also enlarges the attack surface. AIoMT components are exposed to impersonation, replay, and man-in-the-middle attacks. Therefore, secure data exchange becomes essential. Authentication and key exchange (AKE) schemes address this requirement by enabling mutual authentication and session key establishment over insecure channels. Nevertheless, many existing centralized designs suffer from single points of failure and insider threats. Several blockchain-assisted approaches still retain centralized identity traceability. In addition, most AKE schemes either neglect physical security, lack tolerance to intrinsic physical unclonable function (PUF) noise, or store sensitive PUF challenge–response pairs, which increases the risk of modeling attacks. To address these issues, we propose a fully decentralized authentication and key exchange scheme (FDAKES) for AIoMT. FDAKES adopts a$(t,n)$threshold-based root of trust across multiple registration centers (MRCs) to remove unilateral control in registration and tracing. Its server-independent AKE process combines threshold-protected identities, dynamic nonces, timestamps, and PUF and biometric enhanced credentials to achieve perfect forward secrecy. Decentralized conditional traceability preserves user anonymity while allowing identity recovery only with unanimous MRCs consent. By integrating PUF with a fuzzy extractor, FDAKES enables stable secret regeneration without storing raw challenge–response pairs, thereby mitigating modeling threats. We formally prove protocol correctness for login authentication and mutual key agreement. We further establish semantic security of the session key under the real-or-random model, showing that the adversary advantage is negligible in the random oracle model. An extensive informal analysis demonstrates resistance to impersonation, replay, guessing, modeling, physical, man-in-the-middle, and key compromise attacks. Experimental evaluation demonstrates that FDAKES reduces total computational overhead by up to 40.95% and at least 17.25% percent compared with recent state-of-the-art AKE schemes, while communication cost is reduced by up to 76.73% and at least 5% across representative baselines. This work establishes a robust and fully decentralized trust foundation for next-generation smart healthcare systems.

The Fata morgana algorithm (FATA), with its combination of virtual and real search mechanisms and outstanding global optimization ability, has demonstrated strong performance in various optimization problems. This paper attempts to introduce it into the field of feature selection and proposes an enhanced version, named EFATA, which effectively improves the ability to escape local optima and global search by incorporating a collaborative search strategy inspired by swarm foraging. In the experiments, the Support Vector Machine (SVM) is used to guide the feature selection process, enabling the effective identification of key feature across multiple UCI datasets, and experimental results show that it is superior to FATA and other comparison algorithms in terms of classification accuracy and F-score, fully validating the superiority and robustness of the EFATA in feature selection tasks.

Multi-source physiological signals are valuable for quantitative pain assessment, but their high dimensionality, nonlinearity, and strong inter-variable correlations present challenges for clinical decision-making. To improve diagnostic accuracy and efficiency, we propose a novel feature selection framework based on an enhanced Rime optimization algorithm (b-CDRIME), which incorporates a co-adaptive hunting strategy and a dispersed foraging strategy. The framework discretizes the original continuous CDRIME algorithm by introducing a V-shaped binary coding strategy, and constructs a multi-objective weighted fitness function that integrates classification accuracy, feature dimension and AUC, which is combined with the dynamic feedback mechanism of the classifier to achieve efficient feature screening and classification performance optimization. On a multi-source physiological signal public pain dataset, this paper utilizes 10 classical classifiers to test our methods on feature selection and performance improvement. The comparative experiments results show that the pain assessment model constructed based on CDRIME with XGBoost has an average classification accuracy of 81.6% while significantly reducing feature dimensionality. These findings validate the good performance and application potential of this study in the task of processing high-dimensional multi-source pain physiological signals.

Cardiovascular disease (CVD) remains one of the leading global health challenges, calling for accurate and scalable detection methods. This work presents MoViE, a multimodal fusion model based on sparsely activated experts, designed to effectively fuse electrocardiogram (ECG) and electronic health record (EHR) data for CVD detection. MoViE adopts a dual-branch architecture. In the ECG branch, the feed-forward layers of the Vision Transformer (ViT) are replaced with a Mixture-Of-Experts (MoE) module that incorporates top-k routing and shared experts, enabling the model to capture both fine-grained waveform features and long-range temporal dependencies. The EHR branch encodes structured clinical records using TF-IDF, followed by a multilayer perceptron (MLP) to extract semantic representations. To support modality interaction, a MoE-based fusion module is introduced, where a gating network adaptively selects experts to combine complementary features. Extensive experiments on the MIMIC-IV-ECG dataset for myocardial infarction detection demonstrate that MoViE outperforms existing unimodal and multimodal mainstream baselines, achieving 87.98% Accuracy, 90.70% AUC, 66.15% F1-score, and 59.12% MCC, highlighting the potential of MoViE as a general framework for intelligent disease detection.

The use of FPGA technology is becoming more popular for integrating neuromorphic computing systems because of the parallel processing capabilities and flexibility it offers. This study investigates the implementation of the 3D-Morris Lecar neuronal system, known as Digital Optimized Morris Lecar (DOML), in circuits to characterize the magnetic induction flow induced by neuron membrane potential. By employing a combination of trigonometric functions, hyperbolic functions, power-2 based terms, and LUT-based modules, a high-performance circuit is achieved through the best approximation methods. The model’s effectiveness is verified through validation techniques and error analysis, demonstrating a low-error mechanism due to the approximation of nonlinear functions and the avoidance of multipliers, dividers, and high-cost terms. By utilizing the proposed DOML method in digital synthesized circuits, a potential reduction of up to 37% in FPGA hardware resource cost and a potential speed increase of up to 5 times are achievable. The models were digitally synthesized using the Xilinx FPGA Virtex-4 board for cost-frequency validation, showing that the proposed model outperforms the original ML3D model while preserving its essential characteristics. The novelty of this approach lies in applying the combined approximation methods to create a cost-effective circuit suitable for biological systems, offering high speed and low cost attributes. Then, this basic circuit is applied in a sample Network of DOMLs (as a case study) to simulate and realize the population approach in simple form. This research’s findings can have practical implications for the development of neuromorphic hardware in medical devices, particularly in brain-computer interfaces and neuroprosthetics, where high-performance, low-cost hardware is crucial. In large-scale networks, our proposed modeling can be applied which is investigated in case of DOML connections.

The fusion of hyperspectral and multispectral images involves integrating high spectral resolution hyperspectral image (HSI) and high spatial resolution multispectral image (MSI) to generate a HSI with high spatial and spectral resolution (HR-HSI). Existing HSI-MSI fusion methods primarily focus on information fusion within the spatial domain; however, few solutions have explored the employment of frequency analysis to enhance spatial resolution, limiting their capability for global perception. In this paper, we propose an efficient and novel paradigm for HSI-MSI fusion through the cross-frequency enhanced spatial-spectral fusion network, named CESFusion, exploring the complementary fusion of information between the spatial and frequency domains. Specifically, we first present the cross-frequency domain fusion module (CFFM) to perform global analysis through the Fourier transform and effectively integrate and enhance the frequency domain information from both HSI and MSI. Subsequently, we propose the spectral modeling module (SpeMM) based on state space model (SMM) to capture long-range spectral dependencies with linear complexity, and integrate it with the spatial residual block-based module (SRM) for joint spatial-spectral feature extraction. Finally, to enable sufficient interaction between the spatial and frequency domains, we adopt the cross-domain interaction module (CDIM), capturing and integrating complementary information from both domains. Moreover, a frequency-based loss function is purposely designed to further improve the restoration of global information. Extensive experiments conducted on both synthetic and real datasets demonstrate the superiority of our CESFusion, as evidenced by both quantitative and qualitative evaluation results.

Effective discrepancy representation of anomalies plays a crucial role in visual anomaly detection. Recent advances build upon reverse distillation paradigm that boost the teacher–student model’s discrimination capability on anomalies; however, they are still susceptible to the size variation of unpredictable anomalies. To generalize the anomaly size variation, we propose a new algorithm cross-scale denoising reverse distillation (CDRD), which integrates cross-scale denoising with reverse distillation to exchange multiscale perception and enhance the fine-grained representation of features. Specifically, we introduce a cross-scale anomalous signal suppression procedure in the teacher network to facilitate the interaction of information across different scales, thereby enabling the student network to learn more robust normal data representations. In the knowledge transfer process, a fusion compression module acts as an intermediate transmitter of information, aiming to obtain a compact embedding while abandoning anomaly perturbations. Moreover, we construct a detail supplement module in the student network to prevent the loss of key information in the deconvolution process of the decoder. Experiments on well-known datasets demonstrate that our CDRD brings significant improvements over the next best competitor.

With the advent of the Industry 4.0 era and the increasing performance requirements for AR/VR applications and vision assistance and inspection systems in recent years, visual simultaneous localization and mapping (vSLAM) is a fundamental task in computer vision and robotics. However, traditional vSLAM systems are limited by the camera’s narrow field-of-view, resulting in challenges such as sparse feature distribution and lack of dense depth information. To overcome these limitations, this paper proposes a 360ORB-SLAM system for panoramic images that combines with a depth completion network. The system extracts feature points from the panoramic image, utilizes a panoramic triangulation module to generate sparse depth information, and employs a depth completion network to obtain a dense panoramic depth map. Experimental results on our novel panoramic dataset constructed based on Carla demonstrate that the proposed method achieves superior scale accuracy compared to existing monocular SLAM methods and effectively addresses the challenges of feature association and scale ambiguity. The integration of the depth completion network enhances system stability and mitigates the impact of dynamic elements on SLAM performance.

Deep learning has made significant advancements and breakthroughs in medical image recognition. However, the clinical reality is complex and multifaceted, with patients often suffering from multiple intertwined diseases, not all of which are equally common, leading to medical datasets that are frequently characterized by multi-labels and a long-tailed distribution. In this paper, we propose a method involving label decoupling and reconstruction (LDRNet) to address these two specific challenges. The label decoupling utilizes the fusion of semantic information from both categories and images to capture the class-aware features across different labels. This process not only integrates semantic information from labels and images to improve the model's ability to recognize diseases, but also captures comprehensive features across various labels to facilitate a deeper understanding of disease characteristics within the dataset. Following this, our label reconstruction method uses the class-aware features to reconstruct the label distribution. This step generates a diverse array of virtual features for tail categories, promoting unbiased learning for the classifier and significantly enhancing the model's generalization ability and robustness. Extensive experiments conducted on three multi-label long-tailed medical image datasets, including the Axial Spondyloarthritis Dataset, NIH Chest X-ray 14 Dataset, and ODIR-5K Dataset, have demonstrated that our approach achieves state-of-the-art performance, showcasing its effectiveness in handling the complexities associated with multi-label and long-tailed distributions in medical image recognition.

Locating lesions is the primary goal of colonoscopy examinations.3D perception techniques can enhance the accuracy of lesion localization by restoring 3D spatial information of the colon. However, existing methods focus on the local depth estimation of a single frame and neglect the precise global positioning of the colonoscope, thus failing to provide the accurate 3D location of lesions. The root causes of this shortfall is twofold: Firstly, existing methods treat colon depth and colonoscope pose estimation as independent tasks or design them as parallel sub-task branches. Secondly, the light source in the colon environment moves with the colonoscope, leading to brightness fluctuations among continuous frame images. To address these two issues, we propose ColVO, a novel deep learning-based Visual Odometry framework, which can continuously estimate colon depth and colonoscopic pose using two key components: a deep couple strategy for depth and pose estimation (DCDP) and a light consistent calibration mechanism (LCC). DCDP utilization of multimodal fusion and loss function constraints to couple depth and pose estimation modes ensure seamless alignment of geometric projections between consecutive frames. Meanwhile, LCC accounts for brightness variations by recalibrating the luminosity values of adjacent frames, enhancing ColVO's robustness. A comprehensive evaluation of ColVO on colon odometry benchmarks reveals its superiority over state-of-the-art methods in depth and pose estimation. We also demonstrate two valuable applications: immediate polyp localization and complete 3D reconstruction of the intestine. The code for ColVO is available at https://github.com/HNUicda/CoIVO.

The Central Nervous System (CNS) acts as the main element of the biological system, regulating and commanding numerous organs in the human body. Neurons play a crucial role in the central nervous system, and it is necessary to thoroughly examine, replicate, simulate, and integrate various aspects of the CNS to develop a comprehensive neuronal system that can mimic the actual nervous system. In this research, a neuron model called the Quartic Neuron Model is employed to imitate the fundamental nervous functions of the human brain. The proposed method, known as Digital-QNM (D-QNM) is accomplished by employing power-2 based approximation and linear approaches to modify the fourth-degree function. These power-2 based functions are digital-friendly terms (high-accurate, low-cost and leads to high-frequency implementation). By eliminating the high-cost function, the presented model offers advantages such as low error, high speed, and efficient resource utilization compared to the basic main state. In order to validate the final hardware design, a digital FPGA board (specifically, the Xilinx Virtex-5 FPGA board) is employed. The process of digitally synthesizing the hardware demonstrates that our proposed approach can replicate the QNM with improved frequency, performance, and reduced hardware costs. The implementation outcomes show a significant reduction of 98% in FPGA resources cost and a higher operating frequency of the suggested model, reaching 190 MHz. This frequency is considerably higher than the original model’s 105 MHz.

With the advent of the era of Industry 4.0 and the continuous development of point cloud data acquisition technology, point cloud data has been widely used in unmanned distribution of intelligent logistics. This paper designs a 3D point cloud classification model with coordinate attention, blueprint separation involution neural network (BICANet). Firstly, the combination of 2D features and 3D features is adopted to maintain the spatial structure of the point cloud. Secondly, the Involution network is introduced to reduce the amount of redundant data for neural network computation and improve the whole network computation efficiency. At the same time, to further enhance the network feature learning capability, the blueprint separation convolution is combined with coordinate attention. The experimental results prove that the overall accuracy of BICANet in Vaihingen and GML B datasets reaches 86.0% and 98.8%, respectively. It is highly competitive with the currently available methods.

Doctors perform limited one-way intestine endoscopy, in which advanced surgical robots with depth sensors, such as stereo and ToF endoscopes, can only provide sparse and incomplete depth information. However, dense, accurate and instant depth estimation during endoscopy is vital for doctors to judge the 3D location and shape of intestinal tissues, which affects the human-robot interaction between doctors and surgical robots, such as the operation on the subsequent moving of the probe. In this paper, we present a deep learning-based dense depth completion method for intestine endoscopy. We utilize the scattered depth information from depth sensors to make up for the deficiency of features in the intestine and design a multi-scale confidence prediction network to extract dense geometric depth features. Then, we introduce the structure awareness module based on the self-attention mechanism in the depth completion network to enhance the geometry and texture features of the intestine. We also present a virtual multi-modal RGBD intestine dataset and conduct comprehensive experiments on a total of three intestine datasets. The experimental results clearly demonstrate that our method achieves better results in all metrics in all intestinal environments compared to state-of-the-art methods.

Biological systems in case of real-time state and also large-scale simulation approach are interesting and challenge-based due to different aspects of nonlinear mathematical modeling that can describe the interactions of biological blocks. Thus, hardware circuit designing of these basic blocks in the Central Nervous System (CNS) can be an important field in case of achieving high performance neuromorphic system emulator. This paper presents a high-speed, low-cost, and efficient digital circuit for emulating the plausible calcium-dynamic-based model of astrocyte which has spontaneous oscillations. The nonlinear high-cost functions of the complex astrocyte model are reformulated using the power-2 based low-cost terms using optimized exhaustive search algorithm. Subsequently, the proposed model is simulated in case of validating the presented model and new optimized functions. Finally, the proposed model is physically realized in hardware case using Virtex 4 FPGA platform to test and validate final circuits. FPGA implementation results confirmed the ability of the design to emulate biological cell behaviours in detail with high accuracy. The proposed hardware consumes maximum 2% of the all resources of a Virtex 4 board. Additionally, timing analysis and synthesize report represent that the proposed model works in a high frequency of 371.56 MHz. Moreover, to validate the results of implementation, the proposed model is compared with the original model and other similar works in terms of accuracy, speed-up, and maximum number of implemented astrocyte.

In industry 4.0-related applications such as UAVs, autonomous driving, remote sensing, and navigation, environmental perception based on large-scene point clouds plays a crucial role. Accurate point clouds classification is the key and premise of environment perception. In this paper, we propose a new point clouds classification method, FeatureB2SE. First, we design a feature extraction method for point clouds by projecting features in different directions in 2D and 3D to form feature maps. Then, we present a B2SE convolution that can more adequately leverage the advantages from both blueprints separable convolution and Squeeze-and-Excitation networks. To effectively evaluate the performance of FeatureB2SE, extensive experiments have been conducted on two public datasets, GML_B and Vaihingen. The outcome demonstrates that our strategy has achieved state-of-the-art baselines. Specifically, the classification accuracy achieves 98.91% on the GML_B dataset and 85.11% on the Vaihingen dataset, respectively.

With the rapid development of computer computing power and the severe challenges brought by the COVID-19, e-learning, as the optimal solution for most students and other learner groups, plays an extremely important role in maintaining the normal operation of educational institutions. As the user community continues to expand, it has become increasingly important to guarantee the quality of teaching and learning. One way to ensure the quality of online education is to construct e-learning behavior data to build learning performance predictors. Still, most studies have ignored the intrinsic correlation between e-learning behaviors. Therefore, this study proposes an adaptive feature fusion-based e-learning performance prediction model (SA-FGDEM) relying on the theoretical model of learning behav-ior classification. The experimental results show that the feature space mined by fine-grained differential evolution algorithm and the adaptive feature fusion combined with differential evolution algorithm can support e-learning performance prediction more effectively and is better than the benchmark method.

Deep learning has made remarkable achievements in the object classification of 2D images. However, the 3D point cloud classification task is still an open challenge due to the point cloud being irregular and with a mass of noise. This work proposed a large-scale point cloud processing framework that can improve the accuracy and efficiency of point cloud classification. The proposed method calculates the feature values of the point cloud to construct the point cloud feature images, then inputs them into the Graph-MLP++ network to get the point cloud classification result. GraphMLP++ can achieve 97.8% accuracy in the Oakland dataset. Compared with other methods, the efficiency and accuracy of the result are competitive

In a large-scale epidemic, reducing direct contact among medical personnel, attendants and patients has become a necessary means of epidemic prevention and control. Intelligent vehicles and mobile robots in the hospital environment, such as disinfection vehicles, logistics vehicles, nursing robots, and guiding robots, play an important role in improving the operational efficiency of the medical system and promoting epidemic prevention and governance. Powerful capabilities of environmental spatial perception and reconstruction are the keys to accurate localization, navigation, and obstacle avoidance for intelligent vehicles and autonomous robots in such operations. Omnidirectional perception is becoming increasingly important and proliferative in autonomous vehicles and robots since its wide field of view significantly enhances the perception ability. However, the lack of dense and accurate 360° depth datasets has brought the challenge to the omnidirectional perception. In this paper, we propose a depth-sensing and reconstruction system to address this challenge in the large-scale indoor environment. First, we design an omnidirectional depth completion convolutional neural network model, in which a spherical normalized convolutional and the unit sphere area-based loss are introduced to extract features from cross-modal omnidirectional input with unequal sparsity and deal with the imbalanced data distribution and distortion in the panoramic input. In addition, we present a 3D reconstruction system by integrating our depth completion into omnidirectional localization and dense mapping. We evaluate our method on 360D large-scale indoor datasets and real-world sequences of a challenging hospital scene. Extensive experiments show that the proposed method outperforms the other state-of-the-art (SoTA) approaches in terms of depth completion and 3D reconstruction.