This research explores the automated leaf-based identification of medicinal plants, utilizing machine learning and deep learning techniques to address the crucial need for efficient plant classification. Driven by the vast potential of medicinal plants in pharmaceutical development and healthcare, the study aims to surpass the limitations of existing methodologies through thorough experimentation and comparative analysis. The primary goal is to develop a robust and automated solution for classifying medicinal plants based on leaf morphology. The methodology encompasses acquiring diverse datasets. Specifically, set 1 data is processed by applying resizing, rescaling, saturation adjustment, and noise removal, while Set 2 data is processed by applying resizing, rescaling, saturation adjustment, noise removal, and PCA (Principal Component Analysis). The proposed algorithms include Support Vector Machines (SVM), Convolutional Neural Networks (CNNs), YOLOv8, Vision Transformer (ViT), ResNet, and Artificial Neural Networks (ANN). The study evaluates the efficacy and effectiveness of each algorithm in plant classification using metrics such as accuracy, recall, precision, and F1 score. Notably, the ResNet model achieved 93.8% and 94.8% accuracy in Set 1 and Set 2, respectively. The SVM model demonstrated 56.5% and 56.6% accuracy in Set 1 and Set 2, while the Vision Transformer (ViT) model achieved 84.9% and 74.4% accuracy in Set 1 and Set 2, respectively. The CNN model showcased high accuracy at 96.7% and 94.8% in Set 1 and Set 2, followed closely by the ANN model with 96.7% and 96.6% accuracy. Lastly, the YOLOv8 model achieved 96.0% and 95.1% accuracy in Set 1 and Set 2, respectively. The comparative analysis identifies CNN and ANN as the top-performing algorithms. This research significantly contributes to the advancement of medicinal plant identification, pharmaceutical research, and environmental conservation efforts, emphasizing the potential of deep learning techniques in addressing complex classification tasks.

Profit Forecasting for Daily Pharmaceutical Sales Using Traditional, Shallow, and Deep Neural Networks: A Case Study from Sabha City, Libya

This paper investigates the design of a robust controller for the trajectory tracking issue of an underactuated quadrotor unmanned aerial vehicle (UAV) subject to multiple disturbances. An anti-disturbance control framework is proposed by utilizing extended state observer (ESO) and neural network technology. Firstly, the dynamic model of the quadrotor UAV under wind and payload disturbance is established. To actively estimate the lumped disturbance of the UAV system, an ESO with only one parameter is introduced and the disturbances are transformed into the extended state of the UAV system for estimation. Secondly, an adaptive tracking controller that does not accurately obtain the dynamic model knowledge is constructed based on neural network method, where weights of the network can be automatically adjusted by the developed adaptive law. Then, finite-time convergency is analyzed for the ESO with only one parameter, and the Lyapunov criterion is adopted to verify the uniform ultimate boundedness of the UAV closed-loop system. Finally, various simulations under different scenarios are carried out to demonstrate the superiority and effectiveness of the proposed control strategy. For comparison, linear active disturbance rejection control (LADRC), sliding mode control (SMC), model-free based terminal SMC (MFTSMC), and adaptive fractional-order control (ADFOC) algorithms are introduced. Moreover, the physical experiment is given to validate the practicability of the proposed method.

Accurate prediction of the spectral power distribution of light-emitting diodes in multi-phosphor systems is challenging because of the influence of material composition and operating conditions. This paper proposes a spectral prediction framework that combines a Gaussian mathematical model with an improved residual neural network. First, light-emitting diode samples were fabricated using red and green phosphors, and their spectral power distributions were measured. The Gaussian model was then employed to extract the characteristic parameters from the continuous spectral power distributions, and these parameters were used to construct the corresponding dataset. Based on this dataset, a neural network framework was established to map the phosphor mixing ratio, phosphor-to-silicone ratio, and drive current to the Gaussian parameters. Through systematic comparative and extended validation experiments, it is demonstrated that the coefficient of determination for spectral power distribution reconstructed by the Gaussian mathematical model exceeds 0.99. The proposed improved residual network significantly outperforms baseline residual network and recent state-of-the-art methods, achieving superior predictive accuracy and stability. Furthermore, ablation studies validate the effectiveness of the attention mechanism, while sensitivity analyses and independent dataset evaluations further confirm the robustness and generalization capability of the proposed framework. The proposed model significantly enhances spectral prediction accuracy in multi-phosphor systems and achieves rapid mapping from material composition and electrical parameters to the resulting spectrum. A new modeling framework for customized light-emitting diode spectral design is provided in this study, and theoretical support is offered for the intelligent optimization of healthy-lighting and high-color-rendering light sources.

Antimicrobial peptides (AMPs) are universally found in both intracellular and extracellular settings and have significant antibiotic-resistant bacteria are becoming a bigger problem. In medical laboratories, it has shown notable anti-bacterial effectiveness in treating diabetic foot infections and related issues. New medication development frequently targets (AMPs), which are certainly ensuing components of adaptive immune system. The findings of this research employs deep learning to identify antibiotic activity. Numerous computational methods have been established to detect antimicrobial peptides via deep learning algorithms. We introduced a novel deep learning approach called antimicrobial peptides using Capsule Neural Network (AMP-CapsNet) to precisely forecast them and evaluated its efficacy against deep learning and baseline models. AMPs prediction using capsule neural networks, a type of next generation neural network, to build prediction models. Additionally, we utilized Amino Acid Composition (AAC) for effective features encoded method and as well as dipeptide composition (DPC). Every model underwent independent cross-validation and external testing. The findings indicate that the enhanced AMP-CapsNet deep learning model surpassed its counterparts, achieving an accuracy of 97.29% and an AUC score of 98.91% on the test set using with dipeptide Composition (DPC). The proposed AMP-CapsNet demonstrates superior performance of the testing set achieved accuracy 97.29% score with DPC and accuracy 84.42% score with AAC approach. Consequently, the technique we advocate is anticipated to enhance the accuracy of antimicrobial peptide predictions in the future. By producing powerful peptides for medication development and application, this study advances deep learning-based AMP drug discovery approaches. This finding has important ramifications for how biological data is processed and how pharmacology is calculated.

The experimental determination of eutectic points is a long-established and widely used technique, but it is generally only practical for systems with relatively low melting points. Many modern, promising materials, however, are ultra-refractory, with melting points exceeding 3000K. For these systems, conventional melting experiments become prohibitively expensive and technically challenging. Advanced AI modeling can serve as a powerful precursor to guide successful experimentation in such cases. This work proposes a novel criterion for determining the eutectic point concentration in ultra-refractory alloys. The approach is verified using the Ti-B-C system-the most thoroughly studied three-component refractory system to date. The core of the algorithm is a machine-learning interatomic potential, based on a neural network, which achieves accuracy comparable to abinitio methods. Crucially, the algorithm operates effectively in the liquid phase, eliminating the need for information about the solid alloy's crystalline structure to estimate eutectic points.

Cardiac resynchronization therapy (CRT) is a common intervention for patients with dyssynchronous heart failure, yet approximately one-third of recipients fail to respond, partly due to suboptimal lead placement. Identifying optimal pacing sites remains challenging, largely due to patient-specific anatomical variability and limitations of current individualized planning strategies. In a step toward an in-silico approach, we develop two geometric deep learning models, based on graph neural network (GNN) and geometry-informed neural operator (GINO), to predict activation time maps on left ventricular (LV) geometries in real time. Trained on a large dataset generated from finite-element simulations spanning a wide range of synthetic LV shapes, pacing site configurations, and tissue conductivities, the GINO model outperforms the GNN on synthetic cases (1.38% vs 2.44% error), while both demonstrate comparable performance on real-world LV geometries (GINO: 4.79% vs GNN: 4.07%). Using the trained models, we develop a workflow to identify an optimal pacing site on the LV from a given activation time map and show that both models can robustly recover ground-truth subject-specific parameters from noisy inputs. In conjunction with an interactive web-based interface (https://dcsim.egr.msu.edu/), this study shows potential and motivates future extension toward a clinical decision-support tool for personalized pre-procedural CRT optimization.

ABSTRACT This study explores the fabrication of polymer spur gears using fused deposition modeling (FDM). It evaluates their mechanical and surface properties through experimental methods and develops predictive models using artificial neural networks (ANN). Four different thermoplastic materials: PLA, PETG, ABS, and carbon fiber‐reinforced PLA (Cf/PLA) were employed to produce gear models. Dimensional accuracy, surface roughness, Shore D hardness, and wear performance were analyzed under varying printing conditions. A Taguchi L9 orthogonal array was used to investigate the effects of infill density (ID), layer thickness (LT), and printing speed (PS) on gear quality. PLA exhibited the lowest surface roughness (Ra = 11.14 μm), while Cf/PLA provided the highest Shore D hardness of 84.65, along with the lowest coefficient of friction ( μ = 0.1255) under optimized processing conditions. PETG and ABS showed moderate and relatively consistent performance across the evaluated metrics. Dimensional deviations remained under 2.5% for all materials, with Cf/PLA and PLA yielding the highest dimensional stability. ID was the most dominant factor, contributing up to 65.4% to hardness, 59.1% to surface roughness, and 63.7% to wear resistance, depending on material type. For dimensional accuracy, LT had the highest influence, accounting for up to 54.6% of the variation. These findings indicate that both material choice and process parameters have statistically significant effects on the final gear quality. SEM analysis of worn surfaces revealed distinct wear mechanisms: ABS showed adhesive wear and thermal softening, PLA exhibited brittle fracture and thermal degradation, Cf/PLA displayed fiber pull‐out and matrix cracking, while PETG demonstrated abrasive wear and layer delamination, highlighting material‐specific tribological behaviors under dry sliding conditions. ANN was developed to predict gear performance based on material type and processing parameters. The ANN models demonstrated excellent prediction accuracy, with R 2 values exceeding 0.99 and MAPE as low as 8.97% for hardness, 9.68% for surface roughness, 10.38% for wear, and 11.24% for dimensional accuracy.

A well-known and potentially lethal skin cancer requires prompt detection and diagnosis. Complex spatial linkages and global contextual information in skin lesion photos challenge CNNs and other deep learning methods. Given these restrictions, we present a Hybrid Vision Transformer (ViT) with a Graph Neural Network (GNN) and Region-Adaptive Attention to diagnose skin cancer. The ViT branch captures dermoscopy image global dependencies, whereas the GNN enhances features by exploiting lesions' spatial relationships. Region-Adaptive Attention improves lesion categorization by dynamically improving feature extraction in diagnostically relevant locations. Our paradigm for multi-scale lesion analysis accounts for lesion size, color, and texture changes. Meta-learning methods refine the proposed model to make it generalizable across skin tones and imaging settings. Our model outperformed state-of-the-art deep learning algorithms on benchmark skin cancer datasets. The architecture improves classification accuracy and interpretability, making it a promising clinical dermatology tool.

Recent advancements in machine learning (ML) algorithms and applications have made it possible for ML models to solve complex problems, such as reactor core loading optimization, which represents a multiobjective optimization problem with a high degree of freedom. This study aims to provide a proof of concept for an ML-based core loading optimization scheme aimed at research reactors. As a case study we selected the High Flux Reactor in Petten, the Netherlands. Two optimization algorithms are used in this study: genetic algorithm (GA) and reinforcement learning (RL). The goal is to increase the thermal neutron flux at specific locations in the reactor core while adhering to established safety constraints. The optimization schemes also utilized neural network–based surrogate models to substitute for the computationally intensive core calculations. The surrogate models are used to predict core parameters (such as the neutron flux, control rod position, and heat flux) for any given loading pattern. Our results show that ML-based core loading optimization has the potential to become a viable alternative to the traditional core optimization methods. Both the GA and RL methods were able to generate core loading patterns where the neutron flux was similar in most target locations to the results obtained with the traditional method.

All-atom molecular dynamics (MD) simulations are a standard tool for probing the structural and dynamical properties of biomolecular systems, but their accuracy comes at the cost of high computational demands. To overcome spatial-temporal limitations, implicit models or coarse-graining are often employed, but usually at the expense of reduced accuracy. This limitation is also evident in the Poisson-Boltzmann (PB) mean-field theory, which efficiently captures long-range electrostatics but fails to account for crucial short-range interactions. In this work, we bridge this gap by introducing a graph neural network (GNN) Δ-learning approach trained on the difference between all-atom MD and PB, resulting in DIS-PB (deep implicit solvation model using the PB potential as a prior). DIS-PB, which models solutes and salt ions explicitly by MD while water is coarse-grained out, captures both short-range electrostatic correlations as well as long-range electrostatic interaction tails. Applied to a system of the DNA molecule in 1 moll-1 salt solution, our method reproduces structural properties (NDPs, RDFs, and binding probability patterns) with high fidelity, showing that the GNN-corrected PB can reach the accuracy of all-atom MD at a lower computational cost.

Nail diseases, including such common conditions as fungus, and more serious issues like melanoma, may be important clues to the overall health and require a clear diagnosis to be treated. The purpose of the paper is to create a nail disease detection system based on the advanced machine learning methods, including transfer learning and federated learning. The research seeks to show how machine learning and federated learning can be combined to detect nail disease performance with high accuracy without having to share data. The data include pictures of diverse nail conditions including Acral Lentiginous Melanoma, Onychogryphosis, and Pitting among others that are checked to maintain the quality of data in a uniform manner to facilitate the effective training of the models. The most common feature extraction models are ResNet152V2, DenseNet201, MobileNetV2, and InceptionResNetV2 that produce between 1,280 and 2,048 features per image. These characteristics are then pooled to create a unified feature space of 6,784 dimensions which is further narrowed to five major characteristics with Linear Discriminant Analysis (LDA) to create an efficient form of classification. A range of classification models, including Deep Neural Networks (DNN) and Long Short-Term Memory (LSTM) and Bidirectional LSTM (Bi-LSTM) are compared, with the last one reaching the highest classification accuracy of 91.8%. The federated learning strategy enables the joint training of DL models by different clients to ensure data-privacy and has validation-accuracy rates exceeding 99-percent in both uniformly random and structured data distributions. The proposed federated learning-based models resulted high in both uniformly random and structured data distributions.

Recent innovations have made it possible to produce megalibraries, millions of structurally and compositionally distinct nanoparticles on a chip. These megalibraries yield vast volumes of data that are impossible to analyze manually, necessitating the development of automated tools. In previous work, we created a binary classification machine learning model to select quality nanoparticle images for downstream analysis. In this work, we show that adding a custom image processing step before training can produce significantly higher-performing models in a fraction of the time and make them more robust to different image noise levels and microscope acquisition settings. The image processing pipeline proposed here effectively cleans raw nanoparticle images, enhances key features, and allows us to use much lower resolution images and simpler neural network model architectures. These features result in higher performance and significant cost savings. Experiments demonstrate superior performance relative to baseline, including an 18.2% improvement in recall and a 13.1% increase in accuracy. Given the high cost of downstream analysis, it is critical to minimize false positives, and our best-performing model reaches a precision of 95.9% and a weighted F-score of 95.1% on an unseen test set. Additionally, model training time is reduced from hours to less than a minute. We also show that, using this custom image processing pipeline, model performance is significantly improved at lower pixel resolutions compared to downsizing alone. We expect that adopting this pipeline for AI-driven automated nanoparticle characterization will allow researchers to rapidly and accurately analyze much greater volumes of data, thereby accelerating materials discovery.

Capillary tubes are widely used as essential expansion devices in small refrigeration and air-conditioning systems. Accurate prediction of mass flow rate through adiabatic capillaries is a critical aspect of system design and optimization. While there are currently numerous models capable of predicting mass flow through capillaries, most rely on experimental data containing uncertainties, resulting in suboptimal generalization performance. Unlike previous ANN models and empirical correlations that rely on experimental data, this study addresses this limitation by introducing neural networks based on the homogeneous equilibrium model (HEM) of adiabatic capillaries. Two neural networks—a traditional multi-layer perceptron (MLP) and a deep residual network (ResNet)—are developed using a dataset generated by the HEM. The models are subsequently validated and compared against established models using experimental data for various refrigerants and operating conditions collected from the open literature. The results demonstrate that both neural networks exhibit exceptional generalization ability. The average deviations on the experimental dataset are 5.2% for the MLP and 4.5% for the ResNet, outperforming existing models. Their performance across different refrigerants is stable, with the ResNet demonstrating superior overall performance. Furthermore, the trained neural networks achieve a computational speed substantially superior to that of the HEM.

Acknowledging recent breakthroughs in the context of deep bio-inspired neural networks, several architectural deep network options have been deployed to create intelligent systems. The foundations of convolutional neural networks are influenced by hierarchical processing in the visual cortex. The graph neural networks mimic the communication of biological neurons. Considering these two computation methods, a novel deep ensemble network is used to propose a bio-inspired deep graph network for creating an intelligent supply chain model. An automated smart supply chain helps to create a more agile, resilient and sustainable system. Improving the sustainability of the network plays a key role in the efficiency of the supply chain’s performance. The proposed bio-inspired Chebyshev ensemble graph network (Ch-EGN) is hybrid learning for creating an intelligent supply chain. The functionality of the proposed deep network is assessed on two different databases including SupplyGraph and DataCo for risk administration, enhancing supply chain sustainability, identifying hidden risks and increasing the supply chain’s transparency. An average accuracy of 98.95% is obtained using the proposed network for automatic delivery status prediction. The performance metrics regarding multi-class categorization scenarios of the intelligent supply chain confirm the efficiency of the proposed bio-inspired approach for sustainability and risk management.

ABSTRACT This article marks the pioneering effort to explore the remote state estimation (RSE) issue for reaction‐diffusion neural networks (RDNNs) involving semi‐Markov switching coefficients. In the first place, to effectively capture the complex spatial‐temporal dynamics inherent in neural networks (NNs), a semi‐Markov switching NN model incorporating reaction‐diffusion phenomenon is formulated. Subsequently, to acquire the accurate state information of the semi‐Markov switching RDNNs in the context of a remote communication environment, an outlier‐resistant resilient state estimator (ORSE) is developed, accounting for probabilistic gain fluctuations in the estimator and the presence of measurement outliers. This design aims to enhance the robustness of the remote state estimator. Additionally, a novel asymmetric Lyapunov‐Krasovskii functional (LKF) is constructed to alleviate the positive definiteness constraint and reduce conservatism. Furthermore, by employing the LKF approach, sufficient conditions of the asymptotic stability with prescribed performance for the error system are derived. Ultimately, the feasibility of the proposed method is validated through two numerical simulation examples.

Artificial neural networks are reliable machine learning models that have been applied to a multitude of practical and scientific applications in recent decades. Among these applications, there are examples from the areas of physics, chemistry, medicine, etc. To effectively apply them to these problems, it is necessary to adapt their parameters using optimization techniques. However, in order to be effective, optimization techniques must know the range of values for the parameters of the artificial neural network, so that they can adequately train the artificial neural network. In most cases, this is not possible, as these ranges are also significantly affected by the inputs to the artificial neural network from the objective problem it is called upon to solve. This situation usually results in artificial neural networks becoming trapped in local minima of the error function or, even worse, in the phenomenon of overfitting, where although the training error achieves low values, the artificial neural network exhibits low performance in the corresponding test set. To address this limitation, this work proposes a novel two-stage training approach in which a simulated annealing (SA)-based preprocessing stage is employed to automatically identify optimal parameter value intervals before the application of any optimization method to train the neural network. Unlike similar approaches that rely on fixed or heuristically selected parameter bounds, the proposed preprocessing technique explores the parameter space probabilistically, guided by a temperature-controlled acceptance mechanism that balances global exploration and local refinement. The proposed method has been successfully applied to a wide range of classification and regression problems and comparative results are presented in detail in the present work.

Wind energy plays a critical role in the European Union’s decarbonization strategy, including Romania’s growing renewable energy capacity. This study proposes a deep learning-based method for forecasting hourly wind energy production in Romania using feedforward neural networks (FFNNs) and recurrent neural networks (RNNs), trained on a dataset spanning from 1 January to 31 December 2023. The dataset includes hourly wind energy output data (mean = 850.6 MW, std = 694.0 MW) and 13 meteorological variables (e.g., average wind speed = 4.7 km/h, temperature = 14.4 °C). A total of 1296 models were trained and evaluated, with the best-performing RNN model achieving a coefficient of determination of R2 = 0.9680 and a mean absolute error (MAE) of 81.03 MW. The top three models all exceeded R2 = 0.966, demonstrating strong generalization on unseen data. The models were also validated using two external time intervals outside the training/testing sets, confirming robustness. These results show that deep learning models can provide highly accurate, data-driven predictions of wind energy output, supporting grid stability and informed decision-making amid renewable energy variability.

Control techniques for neural-network-based charging stations (CSs) are attracting attention worldwide. This popularity is due to the emergent need for alternative intelligent and adaptive control solutions for attaining a CS with stabilized power transfer and voltage control at the point of common coupling. This paper demonstrates novel neural-network-based improved virtual synchronous motor (NN-i-VSM) control through the mechanism of the charging voltage feedback in conjunction with a trained neural network model to adaptively produce field excitation (MN) that mimics a virtual flux model. The MN adaptively generates an electromotive force based on the trained NN output to control the rectifying converter response of the CS for power quality enhancement during multiple-CS operation. Simulation results in the scenario of multiple CSs at 750 kW (5 × 150 kW) with varying capacities showed significant improvement in voltage variable tracking capacity of up to 500 V as well as power response overshot reduction and grid voltage response tracking improvement compared with an i-VSM-based CS model. A comprehensive CS efficiency assessment and plant stability analysis, including Bode plot evaluation, further confirmed the superior dynamic response performance and robustness of the NN-i-VSM model over the i-VSM model. The proposed model offers scalable applicability in smart mobility and wireless CS integration, signifying a new control advancement for future generations of multiple-grid-friendly charging infrastructure for penetration of batteries at varying capacities.

As deep neural networks (DNNs) become integral to critical applications, protecting their intellectual property (IP) has become paramount. Neural network watermarking is a technique that embeds unique identifiers into models, asserting ownership and deterring unauthorized use. However, sophisticated attacks can deactivate or remove these watermarks without significantly compromising model performance, undermining current protection strategies. In this paper, we introduce the first method for reactivating deactivated neural network watermarks in altered DNN models without requiring access to the original model parameters or training data. By formulating the reactivation process as an optimization problem, we employ projected gradient descent to identify new trigger inputs that restore the embedded watermark. Regularization techniques are incorporated to ensure these triggers resemble legitimate inputs, enhancing both stealth and practicality. Through experiments on various benchmark datasets and model architectures, we demonstrate the effectiveness of our method against common model alterations, including fine-tuning, pruning, and surrogate model attacks. Our work addresses a critical gap in DNN IP protection, offering a robust and practical solution for watermark reactivation. This empowers model owners to assert their rights even in the face of advanced adversarial tactics.