NB-IoT is a form of network, which is designed to link smart devices that do not require substantial power or transmit high data, making it suitable for things such as sensors and trackers. The issue of security is quite a concern because of the numerous interconnected devices and possible attacks. In this paper, a distributed security architecture utilizing DNN-based node behavioral analysis (DNN-NBA) is proposed, and it detects anomalies while minimizing energy consumption. The system does not use a centralized security system, but more so, it decentralizes security tasks to nodes to allow greater scalability and resilience. This DNN-NBA is a deep neural network with four dense layers with 128 neurons, which is capable of examining patterns of device behavior. The proposed model identifies botnet attacks based on network flow-based characteristics that have the ability to capture the behavior of an NB-IoT node. The N-BaIoT dataset was used to train the DNN-NBA model, and it is represented by files with 115 features, with binary classification labels of either benign or transmission control protocol (TCP) attack. TCP attacks are also divided into Mirai and Bashlite attacks in the multiclass categorization. DNN-NBA is energy efficient, relying on lightweight optimization such as Principal Component Analysis (PCA) and extreme gradient boosting (XGBoost) to implement embedded NB-IoT devices. The concept of adaptive learning applied in this case reduces unnecessary computations and enhances energy efficiency. The suggested approach is confirmed by simulations and demonstrates better attack detection capabilities with less energy use than conventional security measures.
Ransomware is one of the most common types of malware that uses cryptology to attack victims’ computers. The victims then have to pay a ransom to recover encrypted resources. Ransomware is currently one of the most serious threats to individuals and organizations. Therefore, it is essential to detect them before they cause serious problems. Since obfuscation tactics are often used in polymorphic and metamorphic ransomware, it is difficult to detect them before they infect the system. Therefore, one should look for a ransomware detection solution that is resistant to obfuscation techniques. The executable file header includes the fields that define the program structure and can potentially be used to detect ransomware before its execution. Extracting this section of executable files does not require preprocessing or special resources. In addition, changing the structure of the program incorporates changes to the header fields as well. The aims of ransomware and benign programs differ, resulting in discrepancies in parts of their headers. In this paper, we propose a new technique called RansPred to accurately detect ransomware utilizing executable file header bytes. RansPred is designed to determine the desired sample label using the alignment score of the important sections of the header and the weighted vote technique. Our results show that RansPred can detect ransomware with 95.0 https://github.com/FarnoushManavi/RansPred.git .
Network intrusion detection systems (IDS) are required to protect the present-day communication infrastructure against the ever evolving cyberattacks. Nevertheless, most of the available IDS solutions have been struggling with unstable features, multifaceted interaction of features, and high imbalance of classes leading to inaccurate detection of low-frequency attacks. To overcome these drawbacks, the present study introduces a deep learning-based system, SAFARI-IDS (Stability-Aware Feature Refinement and Adaptive Robust Intrusion Detection System) that is aimed at enhancing the accuracy of detection, its robustness, and its computational efficiency. The architecture incorporates four modules such as Stability-Aware Feature Refinement (SAFR) to eliminate the unstable and redundant attributes, Interaction-Driven Representation Learning (IDRL) to learn the nonlinear associations between features, Cost-Sensitive Adaptive Decision Engine (CADE) to improve minority attack detection by adaptive weighting, and Adaptive Threshold Stabilization (ATS) to dynamically adapt decision boundaries. The model was tested on the NSL-KDD dataset which consisted of 148,517 records of 41 features and five classes being normal, DoS, Probe, R2L, and U2R. Refined features were 28 in number. Through the experimentation, it has been demonstrated that the SAFARI-IDS performs better than a number of deep learning models based on IDS, with a 98.7
Direct torque control (DTC) is widely acknowledged as an effective approach for regulating the electromagnetic torque of induction motors (IM). However, the variable switching frequency of conventional DTC results in significant flux and torque ripples and degraded low-speed performance. To address these limitations, this work presents an improved DTC strategy based on artificial neural network (ANN) with a flux-oriented structure for voltage-source inverters. The proposed method replaces the conventional hysteresis comparators and switching table, thereby enabling adaptive regulation of stator flux and electromagnetic torque. Magnetic saturation effects are incorporated to improve performance under nonlinear operating conditions. Simulation and experimental results using a TMS320F28379D DSP platform validate the effectiveness and practical feasibility of the proposed approach.
Electric Vehicle (EV) fleet management maintains a group of EVs, which are used for commercial and organizational purposes. Due to the rapid expansion of EV fleets, intelligent management systems are necessary to optimize routes. Existing approaches lack interaction between multiple fleet vehicles and fail to adapt to dynamic grid conditions. Hence, this paper proposes a Wide Slice Dense Network with Adaptive Artificial Lemming Algorithm (WiSDN_Ada-ALA) for EV fleet route optimization and path finding. At first, the EV system model is simulated, where behaviour of EVs is optimized under certain conditions. Then, the input data are normalized using Stopp Normalization. After normalization, intricate features are selected using the Adaptive Artificial Lemming Algorithm (Ada-ALA). Ada-ALA combines an adaptive strategy and Artificial Lemming Algorithm (ALA). Then, data augmentation is performed by Random Value-Based Oversampling (RVOS). Load prediction is done using a Bi-directional Dilated Long Short-Term Memory network (Bi-DLSTM), and then route optimization and path finding are performed by WiSDN, which combines Wide-Slice Residual Networks (WISeR) and Dense Network (DenseNet). Then, the hyperparameters of WiSDN are trained by Ada-ALA. With K-Fold 9, WiSDN_Ada-ALA obtained a charging cost of 9.407 Dollars per kilowatt-hour (/kWh), a distance of 5.191 km (Km), available power of 57.602 Kilowatt (kW), and Normalized Root Mean Square Error (RMSE) of 0.241.
Skin conditions affect millions globally, impacting physical health and quality of life. The accurate classification of skin lesions is critical in dermatology for timely diagnosis and effective treatment, significantly impacting patient outcomes. This paper presents a novel methodology, termed Modified Deep Joint Segmentation with DCNN-SQN Model, designed at improving the classification accuracy of skin lesions through a multi-step process. At first, the input images are pre-processing by Improved Conv.NADE method to enhance their quality. Subsequently, a Modified Deep Joint segmentation model is developed to accurately delineate lesion boundaries. Features essential for classification, including Multi Texton features, shape attributes, and Statistical features, are then retrieved from the segmented regions. Data augmentation methods are utilized to enlarge the data, followed by classification utilizing a hybrid model that integrates DCNN and SqueezeNet architectures. Additionally, an Improved score level fusion technique is implemented to optimize the integration of outputs from both networks, thereby enhancing overall classification performance. The proposed methodology is implemented and evaluated using Python, with comparisons made against conventional methods across diverse criteria, including accuracy, sensitivity, precision, MCC, FNR, and NPV. The results indicate promising advancements in skin lesion classification, providing a robust framework for early detection and intervention.
This manuscript presents a new computational method for solving a class of distributed-order fractional optimal control problems. The proposed method is based on the Ritz-Galerkin method. To this end, we define fractional order generalized Lucas functions. These functions have three free parameters that can affect the accuracy of the methods designed with them. To implement the desired scheme, we present an extra Caputo pseudo-operational matrix, and using it, we propose an extra distributed-order fractional pseudo-operational matrix. With the help of the Ritz-Galerkin method and the mentioned preliminaries, the considered problem is converted to a system of algebraic equations, which can be solved by Newton’s iterative method. We substantiate the superiority of our approach by rigorously comparing it with existing techniques in the literature, affirming its high efficiency.
This paper introduces an integrated deep learning and formal verification framework for automated deadlock detection and resolution in intelligent concurrent systems. By leveraging Long Short-Term Memory (LSTM) networks trained via Backpropagation Through Time (BPTT), the proposed architecture predicts temporal task-resource conflicts with high accuracy. Petri nets serve as the foundational formalism for modeling task interactions and verifying system correctness. Upon deadlock detection, a novel rebuilding module dynamically adjusts execution paths to restore system operability without manual intervention. The framework is validated through a large-scale bridge control system simulation, demonstrating a significant reduction in both deadlock frequency and resolution latency. The proposed methodology offers an efficient and scalable alternative for ensuring correctness and adaptivity in real-time distributed environments.
Smart electrical systems face complex optimization challenges involving multiple conflicting objectives and stringent operational constraints across dynamic scenarios. Existing methods, such as Non-dominated Sorting Genetic Algorithm III (NSGA-III) and Multi-Objective Evolutionary Algorithm (MOEA)/D-DE, handle each scenario independently, leading to inefficient knowledge reuse, whereas Multifactorial Evolutionary Algorithm II (MFEA-II) suffers from adverse transfer under heterogeneous constraints. To address these limitations, this research proposes Bio-inspired Collaborative-Competitive Multi-Tasking Optimizer (BioCoCo-MTO). This model correlated operational scenarios’ peak load, off-peak, and renewable-uncertainty as linked optimization tasks within a unified framework. A constraint-aware selective knowledge transfer mechanism enables beneficial cross-task information sharing while suppressing negative transfer. A symbiotic-competitive evolutionary strategy adaptively balances co-operation and competition among tasks based on task feasibility and constraint-violation behavior. Experiments on the IEEE Reliability Test System Grid Modernization Lab Consortium (RTS-GMLC) benchmark demonstrate that BioCoCo-MTO achieves a mean feasibility rate of 97.6
Modern network architecture is redesigned by Software Defined Networks (SDN), which separate the control plane and data plane. This SDN configuration provides centralized network control and offers flexible programmability; however, it introduces vulnerabilities, making it highly susceptible to cyber threats. This paper introduces a fusion-based multilayered SDN intrusion detection and mitigation framework called Layer-Based-InSDN Dataset-Intrusion Detection System (LB-ISD-IDS). Initially, data preprocessing is done on ISD using label encoding, Z-Score normalization, Recursive Feature Elimination (RFE), and SHapley Additive exPlanations (SHAP). ISD class imbalance is addressed by applying the K-means Synthetic Minority Over-sampling Technique (SMOTE) balancing strategy, which avoids noise amplification. The hybrid ISD attack classifier employs Temporal Convolutional Network (TCN) to extract sequential ISD attack patterns, a LightGradient Boosting Machine (LightGBM) for decision-tree-based rapid and interpretable ISD attack classification, and TabNet for selective attention-based ISD feature transformation. These machine learning and deep learning models in LB-ID-IDS are fused together by using weighted ensemble learning to generate a final ISD attack prediction vector. The ISD attack detection layer is followed by a mitigation layer, which calculates the ISD attack severity score based on which context-aware countermeasures are assigned. Experimental validation of LB-ISD-IDS on ISD achieves an accuracy of 97.57
Cyber-physical systems (CPS), particularly Robot Operating System (ROS) based robotic platforms are widely deployed in safety critical engineering environments, where reliable monitoring and intrusion detection are essential. The existing intrusion detection approaches for cyber-physical systems remain limited by poor generalization, high false positive rates, and inability to effectively model complex temporal–spatial attack behaviors. This highlight a critical need for an intelligent and adaptive framework that can enhance detection accuracy while maintaining robustness for real-time CPS environments. This paper proposes a compact hybrid deep learning framework that combines a convolutional neural network (CNN) front end for per window feature extraction with temporal encoders including bidirectional Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM) and a lightweight Transformer, supported by a streaming oriented preprocessing pipeline. The preprocessing stage applies median imputation for numerical features, top-K encoding for high cardinality categorical attributes and incremental dimensionality reduction to enable real-time processing. Model robustness is enhanced through curriculum based data augmentation and adversarial training using Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD). The proposed CNN + GRU and CNN + Transformer models are evaluated on the full ROSPaCe dataset using class preserving splits and standard multi class intrusion detection metrics. Both architectures achieve approximately 96
This study investigates how hydrodynamic nonlocality influences electromagnetic wave propagation in magnetoplasmas. The bulk-mode dispersion relation is first derived to quantify the impact of nonlocality on the band structure, including dispersion bands and band gaps. With the Chern numbers of the relevant band gaps in mind, the analysis then focuses on identifying the associated edge modes, their dispersion and unidirectionality. To this end, additional boundary conditions induced by nonlocality at plasma interfaces are formulated and used to derive dispersion relations for surface waves at boundaries with dielectrics and unbiased plasmas. The theoretical and numerical results presented in this work may enlighten future developments in plasma-based topological photonic and plasmonic platforms.
Timely clinical intervention relies on the early as well as precise identification of lung cancer from histopathological images. This research introduces an Enhanced Multilayer Perceptron optimized via Artificial Gorilla Troops Optimization (EMLP-AGTO), designed as a streamlined as well as computationally effective substitute for traditional deep CNN methods. The system combines Wiener filtering to reduce noise as well as Local Tetra-Pattern (LTrP) descriptors to extract distinctive texture signatures from H E-stained slides. AGTO dynamically adjusts MLP parameters to ensure stable convergence as well as improve classification dependability. Tests conducted on the LC25000 dataset show robust as well as reliable effectiveness, attaining 99.89
Agriculture is fundamental to human civilization; however, the increasing prevalence and diversity of plant diseases continue to threaten crop yield and quality. Early detection and accurate severity assessment are, therefore, critical for effective disease management. Although recent advances in computer vision have shown promising results in plant disease diagnosis, real-world deployment remains challenging due to complex backgrounds and varying environmental conditions. To address these challenges, we propose Plant Disease Severity Net (PDSNet), a novel framework for disease severity classification in Pome leaves that integrates advanced segmentation, feature extraction, and classification strategies. Mask R-CNN is employed to accurately isolate and segment leaves from complex backgrounds, ensuring precise region-of-interest extraction. PDSNet incorporates a fused feature extraction approach. First, a novel Shifted-3D Local Ternary Pattern (S-3DLTP) descriptor is introduced to capture discriminative textural patterns associated with disease severity, while Shifted Delta Cepstrum (SDC) further improves robustness and spatial feature dynamics. In parallel, an enhanced EfficientNet-B8 architecture with L2 regularization is utilized to extract deep semantic features. Before applying classification, features are optimized using a hybrid metaheuristic optimization algorithm. Experimental results demonstrate that PDSNet achieves validation accuracies of 90.32
This paper presents the first experimental validation of the optimal current grid approximation (OCGA) applied to the design of sparse wire-grid square and triangular trihedral corner reflectors (TCRs) and aims to develop lightweight sparse structures that retain the scattering properties of their solid and wire-grid counterparts. Based on the method of moments, we analyzed the current distribution to identify and remove wires with minimal electromagnetic contribution, utilizing a grid element elimination tolerance of 20
This paper presents a domain-adaptive self-rewarded generative adversarial network (SRWGAN) for multi-pair bidirectional cross-modality image translation. Cross-modality bidirectional translation remains a challenging problem in AI because it requires learning consistent mappings and maintaining feature and cross-domain consistency across multiple imaging domains. While GAN-based methods are popular for cross-modality translation, employing separate generators ( G ) for each direction increases complexity and limits scalability. Moreover, they struggle to adapt to different domains and to generalize across diverse datasets. In addition, traditional G have limited flexibility and depend heavily on discriminator ( D ) feedback, which often leads to superficial imitation instead of learning accurate modality-specific representations. This work introduces a domain-adaptive SRWGAN with a single generator that leverages a RewardNet to create its own feedback, guiding and improving the translation process. This self-rewarding system gives the G more autonomy, enabling it to improve performance without depending entirely on the D . The model also incorporates a dual extremum activation function (DEAF), helping the D better distinguishes between differences. The model can handle bidirectional translation across a variety of medical imaging types, such as MRI ↔ PET, SRS ↔ H E, CT ↔ MRI, and SPECT ↔ MRI. The model is evaluated on ten metrics, including SSIM and DSSIM, and its self-rewarding mechanism helps better align features across modalities, improving overall translation quality. Additionally, Grad-CAM analysis visualized the key regions of emphasis during translation. This bidirectional cross-modality translation SRWGAN uses a self-rewarding mechanism to boost performance, flexibility, and scalability across diverse imaging datasets.
Accurate and reliable fault diagnosis of three-phase induction motors is crucial for enhancing industrial reliability and reducing unplanned shutdown, particularly in variable frequency drive (VFD)–fed systems where signal distortion is considerable. This paper suggests a hybrid and intelligent framework that fuses vibration and current signal that integrates Discrete Wavelet Transform (DWT), statistical feature extraction, and a Dual-branch one-dimensional Convolutional Neural Network (1D-CNN) for the diagnosis and classification of mechanical and electrical induction motor faults. To isolate fault-sensitive frequency bands while mitigating VFD-induced noise, vibration, and stator current signals are decomposed using a seven-level DWT. Discriminative statistical features are extracted from the selected sub-bands and ranked using Information Gain to reduce feature dimensionality and computational complexity. Using a dual-branch 1D-CNN architecture, the optimized feature set is then classified. The suggested method is validated experimentally under various load and speed conditions through nine motor health states, including healthy operation, bearing faults, stator winding short circuits, and broken rotor bars. The results establish a high classification accuracy of 99.4
In Wind Energy Conversion Systems (WECS), cost reduction and reliable operation is achieved by the adoption of Switched Reluctance Generators (SRG) which provide flexible control, improved fault tolerance and robustness. This work focusses on the SRG for WECS with an efficient speed control usingCascaded Adaptive Neuro Fuzzy Inference System (ANFIS), which generates dynamic speed response with improved transient characteristics. The rectified output from generator side is further enhanced by a novel Coupled Inductor (CL) based Modified Single-Ended Primary-Inductor Converter (SEPIC) controlled by Crow search algorithm (CSA) optimized Proportional Integral (PI) controller. The optimized control delivers enhanced outputs at the converter side with high efficiency and voltage values. The proposed work is simulated in Matlab and is also verified experimentally resulting in a reduced settling time of 1.2s with no overshoot issues. The CL-based Modified SEPIC demonstrates an efficiency of 96.4
Ensuring the integrity and credibility of healthcare image data is critical, as any unauthorised modification may lead to severe consequences. Self-embedding watermarking has emerged as an effective signal processing technique aimed at tamper detection and image recovery, particularly in IoT-driven networks. However, many existing block-based medical image watermarking approaches suffer from degraded image quality and limited recovery performance under severe attacks. In this paper, we propose SEWFIR, a secure self-embedding watermarking framework for image recovery. The watermark is generated directly from the host image using Singular Value Decomposition (SVD)-based feature extraction combined with a Normalized Sum (NS) strategy. For robust error detection and correction, a joint Bose–Chaudhuri–Hocquenghem (BCH) and Cyclic Redundancy Check (CRC) coding scheme is employed. Security over IoT-driven networks is further enhanced using Chen’s chaotic system combined with a pseudorandom sequence generator exhibiting strong chaotic behavior. Experimental evaluation demonstrates that the proposed framework attains an average PSNR exceeding 51 dB for watermarked images, while recovered images achieve PSNR values ranging from 25.70 dB to 37.74 dB, outperforming existing methods under various attack scenarios. Additionally, the proposed framework exhibits reduced computational complexity, making it appropriate for practical IoT-based healthcare imaging applications. Unlike deep learning-based watermarking approaches, the proposed framework provides a lightweight and training-free solution suitable for resource-constrained environments.
Permanent Magnet Synchronous Motors (PMSMs) are widely used in Electric Vehicles (EVs) owing to their high efficacy and power density. Nevertheless, it is vulnerable to electrical, mechanical and thermal faults under dynamic operating conditions. To address the limitation of conventional diagnosis methods, this paper proposes a fault diagnosis system based on Artificial Intelligence (AI) powered Black Winged Kite Optimization Algorithm (BWKOA) with hybrid Neuro-inspired Radial Basis Function Spiking Convolution Network (NeuroRBF-SCNet) for PMSM based EV system. Data preprocessing involves data cleaning, data integration and data transformation for removing noise, outliers, combining heterogeneous sensor signals and ensuring numerical stability. Exploratory Data Analysis (EDA) is performed using univariate and bivariate analysis for understanding feature distribution, correlation and fault sensitive pattern. Feature Engineering based on feature scaling is used for balancing multi-sensor inputs, and reducing bias caused by dominant features. The proposed NeuroRBF-SCNet, learns intricate defect patterns by simulating biological spiking responses and Radial Basis Functions (RBF) layer for enhancing nonlinear fault diagnosis. Inspired by the strategic hunting behavior of black-winged kites, a BWKOA is included for dynamic hyperparameter adjustment to improve the system’s learning accuracy. This ensures superior global convergence and avoidance of local optima. The proposed framework is implemented using Python and evaluated on the New Energy Vehicle Diagnosis Dataset across four operating conditions. Experimental results demonstrate the proposed model attaining superior performance with accuracy, recall, F1-Score of 99.09
