Accurate identification of node intrusion behavior in computer networks remains challenging due to the highly dynamic and complex nature of modern network environments, where benign activities are frequently misclassified as malicious events. To address the degradation in detection reliability resulting from such misjudgments, an intrusion identification framework based on an enhanced Recursive Residual Network (RRN) was developed. A statistical classification paradigm was incorporated to process the heterogeneous characteristics of network node intrusion data, enabling a more robust separation of normal and anomalous activity patterns. Features associated with abnormal nodes were extracted, and the range of intrusion-related behavioral deviations was optimized iteratively through an error-minimization function, allowing the model to adapt effectively to fluctuations in network states. A recursive structure derived from Recurrent Neural Network (RNN) principles was subsequently embedded within the residual regression architecture, through which node credit values were continuously iterated and updated to refine the distinction between legitimate behavior and genuine intrusion attempts, enhancing the stability of intrusion identification. In the final stage, a potential loss metric was computed to quantify the expected impact of detected anomalous behaviors on network assets, thereby enabling abnormal behaviors to be classified rigorously as intrusion events when their estimated loss exceeds a critical threshold. Experimental results demonstrate that the proposed method achieves high sensitivity, maintains a stable attack-type identification rate throughout the evaluation period, and reduces the trust value of compromised nodes below 0.15 within 75 seconds, indicating strong effectiveness in distinguishing authentic intrusion behaviors from normal variations. The overall findings suggest that the enhanced RRN offers a resilient and adaptive mechanism for intrusion behavior identification under conditions of complex network dynamics.
Potato production is critically influenced by foliar diseases such as Early Blight and Late Blight, which continue to threaten global food security. Although visual inspection remains widely used, such assessments are subjective, time-consuming, and difficult to scale, creating a pressing need for automated and reliable diagnostic frameworks. In this study, the classification performance and computational efficiency of four state-of-the-art Convolutional Neural Network (CNN) architectures—the Residual Network with 50 layers (ResNet-50), Densely Connected Network with 169 layers (DenseNet-169), EfficientNetV2-B3, and InceptionV3—were systematically benchmarked for the identification of healthy potato leaves and those affected by Early Blight or Late Blight using the publicly available PlantVillage dataset. Accuracy, precision, recall, and F1 score were employed to characterize predictive performance, while parameter count and giga floating-point operations per second (GFLOPS) were used to assess computational efficiency. High-level classification capability was consistently achieved across all models, with overall accuracies ranging from 98% to 99%. DenseNet-169 achieved the highest classification accuracy at 99% with fewer than 13 million parameters, and EfficientNetV2-B3 attained 98% accuracy while exhibiting tsshe lowest GFLOPS requirement. The results indicate that architectures designed for parameter efficiency and feature reuse, such as DenseNet-169 and EfficientNetV2-B3, provide accuracy that is comparable to or surpasses that of less efficient baseline models while offering significant advantages in resource efficiency. These findings reinforce the strong potential of lightweight and high-performance CNN architectures to support scalable, real-time agricultural disease diagnostic systems, particularly in regions where computational resources and technical expertise may be limited.
The long-term resilience of classical cryptographic systems has been challenged by recent advances in quantum computing, particularly through algorithms capable of undermining number-theoretic security assumptions. In this context, a simulation-driven comparative evaluation of Rivest-Shamir-Adleman (RSA) and the BB84 Quantum Key Distribution (QKD) protocol was conducted to elucidate their respective computational, physical, and practical security characteristics. RSA was assessed using OpenSSL implementations across key sizes ranging from 1024 to 4096 bits, with performance quantified through processing time and CPU utilization under controlled experimental conditions. A 31-fold increase in RSA key generation time was observed when scaling from 1024-bit to 4096-bit keys, although overall performance remained compatible with conventional hardware and existing communication infrastructures. In contrast, BB84 was examined using the Qiskit and NetSquid simulation frameworks to analyze photon transmission distance, channel noise, and Quantum Bit Error Rate (QBER) dynamics. The results demonstrate that BB84’s security arises from quantum mechanical principles, with QBER increasing linearly as eavesdropping probability was varied. The comparative analysis reveals that RSA continues to provide practical advantages in software compatibility and computational efficiency. Conversely, BB84 offers a quantum-resistant framework suitable for long-term secure communication. These findings suggest that sustainable cryptographic security is most effectively achieved through hybrid architectures that integrate classical and quantum paradigms, enabling near-term operational feasibility while ensuring future-proof protection against quantum adversaries.
Neuronal oscillations generated by nonlinear dynamical systems have attracted increasing attention for bio-inspired actuation and control applications. In this study, a direct electromechanical coupling between a direct current (DC) motor and neuronal signals produced by a fractional-order Hindmarsh–Rose oscillator (FOHRO) was investigated. The FOHRO was realized using an Arduino–Simulink interface and was employed as a signal-generation unit whose output voltage was used to drive a DC motor. The oscillator was treated as a nonlinear wave-shaping element capable of generating pulse-like and bursting neuronal patterns through appropriate variation of its fractional order and system parameters. The resulting electromechanical system (EMS) was modeled by consistently incorporating Newtonian rotational dynamics and electrical circuit laws. A total energy function was defined, and a scaling transformation was applied to derive an equivalent dimensionless dynamical model. Numerical simulations demonstrated that, when driven by FOHRO-generated neuronal signals, the DC motor exhibited angular velocity responses that preserve the temporal characteristics of the underlying neuronal oscillations, including spiking and bursting regimes. These findings were validated experimentally through real-time microcontroller implementation, confirming close qualitative agreement between simulation and hardware results. The proposed framework provides fundamental insights into the interaction between fractional-order neural oscillators and electromechanical actuators and suggests potential design principles for bio-inspired robotic joints and artificial articulations subjected to electrical stimulation. Such architectures may be particularly relevant for soft robotics, neuro-robotic interfaces, and adaptive actuation systems requiring rich dynamical responses derived from biologically inspired signal sources.
Nonlinear and inherently unstable systems continue to present significant difficulties for conventional control strategies, especially when actuator limitations and parameter uncertainty are involved. This work examines an intelligent control framework in which evolutionary optimization is combined with fuzzy logic control (FLC) to improve the stabilization and robustness of a nonlinear inverted pendulum system. A Mamdani-type fuzzy logic controller is employed to represent nonlinear feedback behavior without resorting to local linearization, while a genetic algorithm (GA) is used to tune the main scaling parameters through simulation-based optimization. The optimization procedure explicitly accounts for nonlinear system dynamics, actuator saturation, and failure-related performance penalties, allowing the evolutionary search to adjust the closed-loop response beyond heuristic parameter selection. Nonlinear time-domain simulations indicate that the optimized controller achieves faster convergence, reduced oscillatory motion, and more consistent performance than a manually tuned baseline controller. Further evaluations under different initial conditions and parameter variations demonstrate an enlarged region of attraction and stable behavior across a range of operating scenarios. These results suggest that evolutionary optimization can play an effective role in shaping fuzzy control structures for nonlinear systems by embedding robustness and performance objectives at the system level. The proposed approach offers a flexible and general framework for the intelligent stabilization of nonlinear and unstable dynamical systems and may be extended to other engineering applications characterized by strong nonlinearity and uncertainty.
Open Access
Journal Metrics
Journal Overview
Associated Journals