The accurate diagnosis of neurodegenerative diseases (NDDs), such as Amyotrophic Lateral Sclerosis (ALS), Huntington's Disease (HD), and Parkinson's Disease (PD), remains a clinical challenge due to the complexity and subtlety of gait abnormalities. This paper proposes the Dual-Branch Attention-Enhanced Residual Network (DAERN), a novel deep learning architecture that integrates Dilated Causal Convolutions (DCCBlock) for local gait pattern extraction and Multi-Head Self-Attention (MHSA) for long-range dependency modeling. A CrossAttention Fusion module enhances feature integration, while SHapley Additive exPlanations (SHAP) and Integrated Gradients (IG) improve interpretability, providing clinically relevant insights into gait-based NDD classification. Uniform Manifold Approximation and Projection (UMAP) visualizations reveal well-separated clusters corresponding to distinct NDDs categories, demonstrating the model's ability to capture discriminative features. Comprehensive ablation studies validate the contributions of model components and preprocessing strategies, highlighting the significance of each in achieving state-of-the-art classification performance. Experimental evaluations on the Gait in Neurodegenerative Disease (GaitNDD) dataset demonstrate that DAERN achieves an accuracy of 99.64%, an F1-score of 99.65%, and an AUC of 0.9997, significantly outperforming conventional deep learning and machine learning baselines. These findings suggest that DAERN could be a valuable and interpretable tool for clinical gait assessment, aiding in early-stage monitoring and automated screening of NDDs, with potential applications in real-time wearable sensor-based gait analysis.

The accurate classification of electrocardiogram (ECG) signals is crucial for the early detection and diagnosis of cardiovascular diseases (CVDs), which remain the leading cause of mortality globally. Traditional methods of ECG interpretation are often limited by their inability to simultaneously identify multiple co-occurring heart conditions, a challenge that is further complicated by the complex nature of ECG signals. In response to this, we introduce an innovative deep learning architecture that combines Efficient Channel Attention (ECA), Squeeze-and-Excitation Networks (SENet), and residual networks (Resnet) modules, specifically designed for multi-label ECG classification using the PTB-XL dataset. Our model not only excels in detecting and distinguishing between various cardiac abnormalities but also addresses the critical need for interpretability in clinical applications. By incorporating SHapley Additive exPlanations (SHAP) and Gradient-weighted Class Activation Mapping (Grad-CAM), we provide transparent, explainable insights into the model's decision-making process, ensuring that healthcare professionals can understand and trust the model's predictions. Experimental results on the PTB-XL dataset demonstrate that the proposed model achieves Exact match, Accuracy and F1-score of 0.638, 88.27% and 91.81%, respectively, with substantial improvements in both sensitivity and specificity across various cardiac conditions compared to state-of-the-art methods. This work represents a crucial step toward more reliable, interpretable, and clinically applicable AI-driven diagnostics for cardiovascular health.

Accurate and robust fault diagnosis of rolling bearings under varying operational conditions is critical to ensuring the reliability and safety of industrial systems. However, domain shifts caused by changes in load, rotational speed, sensor configurations, and data acquisition platforms often lead to significant degradation in diagnostic performance. To address this challenge, we propose a novel cross-domain bearing fault diagnosis framework that integrates a multi-loss deep adaptation strategy with confidence-aware pseudo-label refinement and Maximum Mean Discrepancy (MMD)-based distribution alignment. From an artificial intelligence perspective, the architecture comprises three key components: (1) a shared Wide Deep Convolutional Neural Network enhanced with a Bidirectional Gated Recurrent Unit (BiGRU) module for temporal-spatial feature extraction across domains; (2) MMD-based domain alignment minimizing feature distribution discrepancies; and (3) confidence-aware pseudo-labeling selectively incorporating high-confidence target predictions. These are jointly optimized using a unified loss integrating supervised classification, unsupervised MMD alignment, and self-supervised pseudo-label refinement. Extensive experiments on Paderborn University and Jawaharlal Nehru University datasets demonstrate superior performance across multiple transfer scenarios. The method achieves 94.85% and 99.47% cross-domain accuracy respectively, significantly outperforming baselines. Ablation studies validate the complementary roles of the MMD and BiGRU modules. The lightweight design (0.0976M parameters, 2.02M Floating Point Operations Per Second) enables real-time inference (0.927ms/sample), outperforming Residual Neural Network-based methods by 95% in computational cost while maintaining reliability. From an engineering application standpoint, this framework addresses critical industrial challenges in bearing fault diagnosis under varying conditions, providing computationally efficient solutions for real-time deployment in manufacturing and energy systems where labeled target data are unavailable.

Epilepsy is a chronic neurological disorder characterized by recurrent seizures. Accurate diagnosis and effective monitoring require the precise classification of electroencephalogram (EEG) signals. In this study, we introduce a novel approach that combines Adaptive Local Iterative Filtering (ALIF) for signal decomposition with an attention-enhanced cascaded deep neural network (CDNN) architecture. The ALIF algorithm decomposes EEG signals into intrinsic mode functions (IMFs) that capture inherent oscillatory components. These IMFs are processed by the CDNN, which operates in two stages: a feature extraction module and a classification module. In the feature extraction stage, a SEblock channel attention mechanism dynamically prioritizes significant features within the IMFs. The classification stage employs a hybrid CNN-LSTM architecture that effectively captures both spatial and temporal dependencies. To enhance interpretability, the SHapley Additive exPlanations (SHAP) framework is incorporated to provide insights into the model's decision-making process, while Gradient-weighted Class Activation Mapping (Grad-CAM) visualizes the most discriminative regions in the input data. Rigorously validated using 10-fold cross-validation on the Bonn and EEG Epilepsy databases, the proposed methodology achieved an exceptional classification accuracy of 100%, with sensitivity, specificity, and F1-scores exceeding 99% across various scenarios. The integration of SHAP and Grad-CAM not only elucidates the model's decision processes but also contributes to a more interpretable and reliable system for epileptic EEG signal classification. This synergistic combination of advanced signal processing, deep learning, and interpretability techniques holds significant potential to enhance epilepsy diagnosis and strengthen trust in clinical decision support systems.

This paper focuses on the trajectory tracking control of an underactuated AUV in the vertical plane. First, an error analysis is conducted for the underactuated AUV in the vertical plane, and a Disturbance Observer (DO) is designed to estimate external disturbances, while a Radial Basis Function Neural Network (RBFNN) is employed to approximate nonlinear terms. Subsequently, controllers are designed based on Nonsingular Terminal Sliding Mode Control (NTSMC) and Nonsingular Fast Terminal Sliding Mode Control (NFTSMC), respectively. Finally, the stability of the controllers is verified using the Lyapunov function. Simulation results demonstrate that controllers effectively improve trajectory tracking accuracy and enhances the robustness of the system.

Path-following control for wheeled mobile robots operating on varying terrains constitutes a fundamental challenge in robotic motion control, as terrain variations directly impact steering dynamics. This study models steering dynamics as a first-order system, where the dynamic parameter is matched to specific terrains. To achieve terrain adaptation, Support Vector Machine (SVM) methodology enables real-time matching of steering dynamics parameters to identified terrains through instantaneous image recognition. By integrating time-varying steering dynamics with fundamental kinematics, we establish the control model and subsequently design a path-following controller for this time-varying system using Lyapunov stability theory. Comparative experiments conducted in variable terrain environments evaluated fixed-parameter versus terrain-adaptive parameter control across linear, square, and circular paths. The results demonstrate that when the dynamic parameter can adapt to terrain variations, the designed time-varying parameter path following controller enables the mobile robot to converge to the predefined path more rapidly with reduced overshoot.

Path following presents a pivotal challenge within the realm of small fixed-wing unmanned aerial vehicles. Firstly, a Lyapunov-stable path guidance law was formulated to follow specific planar curved paths. To ensure differentiability of the guidance law, a modified, smooth saturation function was derived. Secondly, an analysis was conducted to ascertain the interrelationship between control parameters and input constraints, thereby identifying the relevant parameter domains. Thirdly, the nonlinear model predictive control technique was harnessed to optimize both guidance law parameters, enhancing the unmanned aerial vehicle’s capacity to achieve optimal performance in both straight-line and circular path following, hereafter referred to as PFC_NMPC. By leveraging Lyapunov stability arguments for switched systems, the stability of the corresponding nonlinear switched system was guaranteed. In this study, square and circular paths were generated to assess the path-following control of a simulated fixed-wing unmanned aerial vehicle. The performance of various guidance laws, including those with fixed parameters (PFC), those with parameters tuned using fuzzy logic (PFC_FL), PFC_NMPC, vector field, and pure pursuit with line-of-sight, was compared. Notably, the proposed PFC_NMPC method exhibited the ability to expedite the unmanned aerial vehicle’s convergence to the desired path while maximizing the effective flight path length.

Trajectory planning plays a crucial role in the execution of Unmanned aerial vehicle (UAV) missions. However, planning an optimal collision-free trajectory is a challenging task, especially in complex environments. To address the above issue, an enhanced Elliptical tangent graph algorithm (ETG-CPI) based on comprehensive performance indicator is proposed in the present study. In the proposed algorithm, the comprehensive performance indicator, which contains the obstacle avoidance frequency, the yaw angle and the distance from the start point to the candidate waypoint, is used to select promising waypoints. Moreover, the entropy weight method is used to integrate these performance indicators. The experimental results demonstrate that the proposed algorithm outperforms four competitive path planning methods in 26 different environments. Additionally, the results indicate that the proposed comprehensive path evaluation method can help the proposed algorithm find a high-quality path in complex environments.

Epilepsy, a neurological disorder marked by recurrent seizures, demands precise classification of electroencephalogram (EEG) signals for effective diagnosis and monitoring. This paper introduces a novel methodology that integrates the Adaptive Local Iterative Filtering (ALIF) for signal decomposition and leverages cascaded deep neural networks for robust signal classification. The motivation arises from the critical need for heightened accuracy in epilepsy-related EEG signal analysis. The research methodology begins with the application of ALIF to decompose EEG signals. This process generates multiple Intrinsic Mode Functions (IMFs) that capture the inherent oscillatory components of the signals. The subsequent step involves inputting these IMFs into a cascaded deep neural network. Within the cascaded deep neural network (CDNN), a feature extraction module utilizes the SEblock channel attention mechanism to automatically extract salient features from the IMFs. This attention mechanism enhances the network’s ability to focus on essential information, improving the overall discriminative power of the model. Following feature extraction, the classification module of the CDNN is employed to categorize the EEG signals into their respective classes. The entire process is subjected to rigorous validation using a 10 -fold cross-validation strategy. The proposed methodology demonstrates exceptional performance when tested on both the Bonn and EEG Epilepsy databases. Achieving a classification accuracy of 100% in both cases highlights the efficacy of the approach in accurately identifying epileptic EEG signals, showcasing its potential for reliable clinical applications.

Deep Learning has been widely used in the monitoring and diagnosis of wind turbines. However, most of the current fault diagnosis methods only use single sensor signal as the input of DL model, which leads to the limitation of the model performance. Therefore, this paper proposes a multi-signal CNN-GRU model. Firstly, the acquired multiple sensor signals are converted to time–frequency images by Multi-Synchrosqueezing S-Transform, the frequency domain features of multiple sensors are extracted by Convolutional Neural Network and fused by Attention Mechanism, then the multi-source time-frequency features are extracted by Gated Recurrent Unit and finally classified by SoftMax. Experiments are conducted on the CWRU dataset and the field gearbox dataset. The results show that the proposed method achieves an average accuracy of 99.69% and 100% on the two datasets, which are both higher than existing DL-based fault diagnosis methods. The proposed method can effectively fuse signals from multiple sensors, thus improving the classification accuracy and stability of the model, which has high practicality and reliability for fault diagnosis of wind turbines.

Accurate real-time kinematics model is very important for the control of a skid-steering mobile robot. In this study, the kinematics model of the skid-steering mobile robots was first designed based on instantaneous rotation centers (ICRs). Then, the extended Kalman filter (EKF) technique was applied to obtain the parameters of ICRs under the same specific terrain online. To adapt to different terrain environments, the fractal dimension-based SFTA (segmentation-based fractal texture analysis) method was used to extract features of different terrains, and the k-nearest neighbor (KNN) method was used to classify the terrains. In the case of real-time terrain recognition, the filter parameters of the EKF for estimating the ICRs are adjusted adaptively. Experiments on a real skid-steering mobile robot show that this method can quickly estimate the kinematics model of the robot in the case of terrain changes, and can meet the needs of practical applications. The average error of odometer estimation based on visual terrain classification is 0.06 m, while the average error of odometer estimation without terrain classification is 0.14 m.

Our study proposes an automatic technique for the detection of MI using hybrid signal processing tools and deterministic learning theory. To begin with, the characteristic envelope and first derivative of a single-lead ECG signal are extracted using Shannon energy. By incorporating the Shannon energy envelope (SEE) in the phase portrait of an ECG signal, the non-linear system dynamics can be captured. In the second step, using fast and adaptive multivariate empirical mode decomposition (FA-MVEMD) with the SEE of the ECG and its derivative, scale-aligned intrinsic mode components (IMFs) are generated. The two initial IMFs, which contain the most energy in the ECG signals and their derivatives, are considered the predominant IMFs and employed as features. There are significant differences between ECG signals produced by normal (healthy) and MI-related cardiac systems. The third step involves applying deterministic learning theory with neural networks to model, identify and classify ECG signals into two groups. As a final step, an evaluation of the effectiveness of the method is then performed on the PTB diagnostic ECG database, which comprises signals from 148 MI patients and 52 healthy controls. The average classification accuracy achieved with a cross-validation scheme of 10 folds is reported as 99.21 %. In conclusion, our results confirm the proposed features are consistent with ECG system dynamics, as well as complementing existing ECG features for automatic MI detection.

In this paper, the problem of horizontal path following of a small fixed-wing unmanned aerial vehicle (UAV) is studied based on the combined kinematic/dynamic model. First, a stable kinematic path following control law is designed. Second, the backstepping technique is applied to derive the roll angle command by considering the closed-loop roll dynamics. The stability of the corresponding nonlinear system is guaranteed by Lyapunov stability arguments. Third, two key parameters of the control law which will influence the path following performance have been analyzed. Two fuzzy logic based controllers have been designed to optimize the parameters to improve the path following performance (PFC_FL), where the stability of the corresponding nonlinear system is still kept. The simulations of square and circular paths following control of an Aerosonde UAV have been studied. The path following performances of the proposed control law with parameter fixed (PFC), PFC_FL method, vector field (VF) method, and pure pursuit with light-of-sight (PLOS) method were compared. The proposed PFC_FL method can make the UAV have the longest length of corresponding effective flight path and the shortest total convergence time. The PFC_FL can also have much smaller error overshoots than those by using the method PFC during route switching.

Understanding and analyzing 2D/3D sensor data is crucial for a wide range of machine learning-based applications, including object detection, scene segmentation, and salient object detection. In this context, interactive object segmentation is a vital task in image editing and medical diagnosis, involving the accurate separation of the target object from its background based on user annotation information. However, existing interactive object segmentation methods struggle to effectively leverage such information to guide object-segmentation models. To address these challenges, this paper proposes an interactive image-segmentation technique for static images based on multi-level semantic fusion. Our method utilizes user-guidance information both inside and outside the target object to segment it from the static image, making it applicable to both 2D and 3D sensor data. The proposed method introduces a cross-stage feature aggregation module, enabling the effective propagation of multi-scale features from previous stages to the current stage. This mechanism prevents the loss of semantic information caused by multiple upsampling and downsampling of the network, allowing the current stage to make better use of semantic information from the previous stage. Additionally, we incorporate a feature channel attention mechanism to address the issue of rough network segmentation edges. This mechanism captures richer feature details from the feature channel level, leading to finer segmentation edges. In the experimental evaluation conducted on the PASCAL Visual Object Classes (VOC) 2012 dataset, our proposed interactive image segmentation method based on multi-level semantic fusion demonstrates an intersection over union (IOU) accuracy approximately 2.1% higher than the currently popular interactive image segmentation method in static images. The comparative analysis highlights the improved performance and effectiveness of our method. Furthermore, our method exhibits potential applications in various fields, including medical imaging and robotics. Its compatibility with other machine learning methods for visual semantic analysis allows for integration into existing workflows. These aspects emphasize the significance of our contributions in advancing interactive image-segmentation techniques and their practical utility in real-world applications.

Fixed-wing vertical take-off and landing (VTOL) UAVs have received more and more attention in recent years, because they have the advantages of both fixed-wing UAVs and rotary-wing UAVs. To meet its large flight envelope, the VTOL UAV needs accurate measurement of airflow parameters, including angle of attack, sideslip angle and speed of incoming flow, in a larger range of angle of attack. However, the traditional devices for the measurement of airflow parameters are unsuitable for large-angle measurement. In addition, their performance is unsatisfactory when the UAV is at low speed. Therefore, for tail-sitter VTOL UAVs, we used a 5-hole pressure probe to measure the pressure of these holes and transformed the pressure data into the airflow parameters required in the flight process using an artificial neural network (ANN) method. Through a series of comparative experiments, we achieved a high-performance neural network. Through the processing and analysis of wind-tunnel-experiment data, we verified the feasibility of the method proposed in this paper, which can make more accurate estimates of airflow parameters within a certain range.

Energy Storage System (ESS) has been utilized extensively to manage the uncertainty of renewable energy output and load demand. To utilize ESS more effectively, the concept of shared energy storage system (SESS) is proposed. This paper proposes an adjustable robust optimization model for microgrid cluster, taking into consideration the uncertain renewable energy and load. The robust adjustment parameter Γ is introduced to adjust the conservative degree of the model. According to the predicted WT and PV outputs, load demand and given Γ, the reserve capacity and pre-scheduling scheme of microgrid cluster are solved. By simulating the actual fluctuation of uncertain parameters and re-scheduling, the robustness and economy of pre-scheduling schemes under different Γ are verified. Simulation results demonstrate that the adjustable robust optimization model is superior to both the deterministic optimization model and the static robust optimization model. Furthermore, comparing the operating costs of each microgrid with ESS and the microgrid cluster with SESS, it is proved that SESS not only reduces the charging and discharging frequency of ESS, but also decreases the requirement for reserve capacity, thereby reducing the operation cost.

This study aims to develop a new approach for detecting OSA automatically with ECG system dynamics disparity. First, tunable quality factor (Q-factor) wavelet transform (TQWT), variational mode decomposition (VMD) and phase space of three-dimension (3D) are combined for the feature extraction, which can carry clinically relevant information about the anomalies present in the OSA ECG recordings. Second, dynamic modeling and identification of ECG systems are carried out using neural networks. By constructing on a bank of dynamical estimators, normal and OSA ECG signals will be classified based on their differences in dynamics. Finally, a physionet apnea-ECG database consisting of 70 overnight recordings from 70 participants is used to evaluate the outcomes. By using a cross-validation scheme of 10 folds, experimental results indicate that neural network based classifier along with proposed features achieves higher accuracy, sensitivity, and specificity values of 98.27%, 97.68%, and 98.63%, respectively. Results validate that the proposed method may serve as an alternative to PSG to detect OSA automatically in clinical settings.

Visual saliency refers to the human’s ability to quickly focus on important parts of their visual field, which is a crucial aspect of image processing, particularly in fields like medical imaging and robotics. Understanding and simulating this mechanism is crucial for solving complex visual problems. In this paper, we propose a salient object detection method based on boundary enhancement, which is applicable to both 2D and 3D sensors data. To address the problem of large-scale variation of salient objects, our method introduces a multi-level feature aggregation module that enhances the expressive ability of fixed-resolution features by utilizing adjacent features to complement each other. Additionally, we propose a multi-scale information extraction module to capture local contextual information at different scales for back-propagated level-by-level features, which allows for better measurement of the composition of the feature map after back-fusion. To tackle the low confidence issue of boundary pixels, we also introduce a boundary extraction module to extract the boundary information of salient regions. This information is then fused with salient target information to further refine the saliency prediction results. During the training process, our method uses a mixed loss function to constrain the model training from two levels: pixels and images. The experimental results demonstrate that our salient target detection method based on boundary enhancement shows good detection effects on targets of different scales, multi-targets, linear targets, and targets in complex scenes. We compare our method with the best method in four conventional datasets and achieve an average improvement of 6.2% on the mean absolute error (MAE) indicators. Overall, our approach shows promise for improving the accuracy and efficiency of salient object detection in a variety of settings, including those involving 2D/3D semantic analysis and reconstruction/inpainting of image/video/point cloud data.

Phonocardiogram (PCG) is commonly used as a diagnostic tool in ambulatory monitoring in order to evaluate cardiac abnormalities and detect cardiovascular diseases. Although cardiac auscultation is widely used for evaluation of cardiac function, the analysis of heart sound signals mostly depends on the clinician’s experience and skills. There is growing demand for automatic and objective heart sound interpretation techniques. The objective of this study is to develop an automatic classification method for anomaly (binary and multi-class) detection of PCG recordings without any segmentation. A deep neural network (DNN) model is used on the raw data during the extraction of the features of the PCG inputs. Deep feature maps obtained from hierarchically placed layers in DNN are fed to various shallow classifiers for the anomaly detection, including support vector classifier (SVC), k-nearest neighbors (KNN), random forest (RF), gradient boosting (GB) classifier, decision tree (DT) classifier, quadratic discriminant analysis (QDA), and multi-layer perception (MLP). Principal component analysis (PCA) technique is used to reduce the high dimensions of feature maps.Finally, two famous heart sound databases, namely PhysioNet/Computing in Cardiology (CinC) Challenge heart sound database and heart valve disease (HVD) database, are used for evaluation. The databases are significantly different in terms of the tools used for data acquisition, clinical protocols, digital storages and signal qualities, making it challenging to process and analyze. By using the 10-fold cross-validation style, experimental results demonstrate that the proposed deep features with shallow classifiers yield highest performance with accuracy of 99.61% and 99.44% for binary and multi-class classification on the two databases, respectively. The results indicate that our method is effective for the detection of abnormal heart sound signals and outperforms other state-of-the-art methods.

Due to the coupling characteristics of the physical system and the communication system, distribution network fault scenarios and post-disaster recovery procedures become more complicated in the aftermath of catastrophic natural disasters. Consideration of communication system restoration by unmanned aerial vehicle base station (UAV-BS) can effectively reduce distribution network outage duration. This paper proposes a post-disaster recovery strategy for distribution networks that takes UAV-based communication recovery into account. Consideration is given to the cooperation of multiple UAV-BSs in the recovery process using multi-agent reinforcement learning (MA-RL). The communication recovery procedure of multiple UAVs is initially converted into a Markov decision process (MDP). To actualize agent interaction, the distribution network reconfiguration model is constructed as a reinforcement learning environment that takes into account communication constraints. The problem is resolved by MA-RL, and the effectiveness of the proposed strategy is evaluated by IEEE 33-bus system.