Arabic handwriting recognition is a very active research area that poses a significant challenge due to the complexity and diversity of the Arabic script. Deep learning techniques are becoming increasingly popular, particularly in computer vision applications. It is a promising approach for improving pattern recognition tasks in general. This study aims to use a convolutional neural network (CNN) and a long short-term memory network (LSTM) for offline recognition of handwritten Arabic characters. A series of experiments was carried out to create an effective model for each neural network and to evaluate and compare the results of these two techniques. A combined model of the two networks was also proposed. The effectiveness of the models is assessed using the AHCD database. The results show an accuracy of 98.57
This study investigates the effect of neural network architecture on the accuracy of data-driven modeling of thermal explosions in a hydrogen-oxygen-air mixture. Using a reduced kinetic mechanism for 11 reagents, the thermal explosion process is simulated under specified initial pressure and temperature conditions, generating time-resolved data. We compare three architectures: a standard multilayer perceptron (MLP), a DeepONet-inspired model, and our U-Net-style residual network, evaluating their ability to capture transient dynamics and key reaction regimes. Our results demonstrate that network architecture has an important impact on predictive performance. Among the three considered architectures, the U-Net-based model demonstrated the lowest prediction error, with mean squared error (MSE) of 1.3 × 10–3 and a standard deviation (STD) of 2.18 × 10–2, indicating stable performance across the test set. By comparison, both the DeepONet-inspired model and the MLP exhibited substantially larger errors, with MSE values of 1.81 × 10–2 (STD 5.81 × 10–2) and 2.02 × 10–2 (STD 6.82 × 10–2), respectively. The increased dispersion of errors for these two models was particularly pronounced in trajectories involving rapid temperature growth. These findings suggest that the careful choice of network architecture is a critical factor in the development of reliable neural-network surrogates for combustion and reactive-flow modeling.
Emotion recognition plays a crucial role in applications such as human-computer interaction, healthcare, adaptive learning, and social media analysis. However, existing multimodal frameworks often struggle to effectively model paralinguistic features and manage the random fusion of different modalities, resulting in misclassifications and limited interpretability. To address these challenges, this paper introduces TriModal-SentiXNet, a novel multimodal framework designed for dual emotion and sentiment recognition using text, speech, and facial expressions. The framework operates in three main stages: pre-processing, feature extraction, and explainable classification. The pre-processing phase employs a Multimodal Adaptive Pre-processing (MAP) method that enhances input quality by denoising speech, standardizing visual frames, and semantically enriching textual data. For feature extraction, the RoBi-FuseNet architecture learns comprehensive contextual representations from each modality while effectively modeling temporal dynamics and cross-modal interactions to improve recognition accuracy. Explainable classification is achieved using SHapley Additive exPlanations (SHAP), which provides transparency by quantifying each modality’s contribution and ensuring balanced influence. The integration of adaptive pre-processing and context-aware fusion strengthens robustness, interpretability, and overall system reliability. Experimental evaluations on benchmark datasets CMU-MOSEI and CMU-MOSI demonstrate notable performance gains, achieving accuracies of 87.92 and 86
Monogenetic disorders result from mutations in a single gene, making early diagnosis essential for preventive healthcare. Timely and accurate identification is crucial for effective clinical intervention, genetic counseling, and personalized treatment. However, the high dimensionality, redundancy, and complexity of genomic data present major challenges for traditional diagnostic models. In this study, a novel hybrid framework of Adaptive Elistic Genetic Algorithm Boosted Feedforward Neural Network for Monogenetic Disorder Diagnosis (AEGA-BoostFNN-MonoDx) Model is proposed for monogenetic disorder prediction. This approach integrates with an Adaptive Elistic Genetic Algorithm (AEGA) with a Boosted Feedforward Neural Network (BoostFNN) to enhance the predictive accuracy. The AEGA module is designed to perform intelligent feature selection using adaptive mutation and elitist strategies. This ensures the identification of highly informative and non-redundant features while minimizing the computational complexity. The optimized features are then used to train the BoostFNN classifier, which captures the complex nonlinear patterns in genomic sequence data. The model is evaluated by using datasets such as the Of Genomes And Genetics Datasets and Genetic Variant Classifications Datasets, encoded using biologically meaningful representations such as k-mer and one-hot encoding. The AEGA-BoostFNN-MonoDx Model outperforms the existing model, achieving 98.75
Cloud computing provides on-demand services with high performance and scalability over the Internet. The primary goals of task scheduling in cloud environments are to efficiently utilize available resources, minimize execution time, and ensure timely task completion. Load balancing is crucial to ensure that virtual machines (VMs) are evenly utilized. However, the key challenge in cloud computing lies in effectively balancing the load and scheduling tasks. In the proposed model, an optimal task-scheduling mechanism is designed using the State-Action-Reward-State-Action (SARSA) algorithm. The scheduling process is enhanced by the Osprey Optimization Algorithm (OOA) to select the best resources, minimizing execution time, cost, and resource utilization. Once the tasks are schelued in queue, the load balancing is carried out using the Walrus Optimization Algorithm (WaOA), which optimally balances the load across VMs based on task lifetime and response time. The performance of the proposed model is evaluated under various task conditions, and the results are compared to existing models for validation. The proposed approach demonstrates superior performance with a makespan time of 64.81 s, turnaround time of 7.05 s, waiting time of 203.48 s, response time of 70 s, scheduling time of 226 s, a success rate of 95.7
The need for automatic solar panel defect detection is becoming more and more urgent due to the rising demand for new solar energy systems worldwide. Convolutional neural networks are very effective at solving the image classification problem across a wide range of domains. A model for detecting faults in solar panel surfaces was presented in this paper. The three primary stages of the suggested model are the data preparation stage, the ResNet50 hyperparameter optimization stage, and the image classification stage. Furthermore, a modified version of the light spectrum optimizer (MLSO) was introduced in this paper. In the optimization process of the original light spectrum optimizer, the proposed MLSO makes use of the outward search method. The ideal batch size and learning rate values are determined using MLSO. A publicly available dataset with six classes is employed. The proposed fault detection model can greatly improve the accuracy of solar panel defect detection, according to the experimental results. The model achieved an accuracy of 96.58
Crop yield prediction algorithms let farmers decide when and what kinds of crops to cultivate based on environmental factors, which increase yields. Predicting crop yields is difficult because of a number of intricate elements, including genotype, water availability, pest infestations, landscape quality, and harvest scheduling. By addressing these challenges through robust methodologies and innovative approaches, machine learning and deep learning models have the potential to greatly improve the accuracy and dependability of crop production forecast, thereby supporting global efforts to ensure food security and sustainable agriculture. This study utilizes Deep Learning and Machine Learning algorithms to analyse crop cultivation stages, providing solutions based on meteorological conditions, fertilizer requirements, and Growing Degree Days. Production in agriculture efficiently uses machine learning algorithms like RF, KNN, MSER, BMA, and SVM for crop yield prediction. BMA’s machine learning model has a high prediction accuracy rate of 98
Osteoporosis (OP) is a prevalent metabolic bone disease characterized by decreased bone density and strength, leading to an increased risk of fractures, especially among the elderly. This condition often remains undiagnosed until a fracture occurs, making early detection crucial to prevent complications, reduce morbidity, and improve the patient’s quality of life. Currently the bone density was to identify the osteoporosis, and Dual Energy X-ray Absorptiometry (DEXA) for figuring out the size of the bone, time consuming, and the manual interpretation having the error. A novel fusion method for radiography images contrast augmentation was developed in order to avoid this issue. Initially, both X-Ray and subject records are considered as the dataset. The images are enhanced for clear bone identification through pre-processing techniques including resizing, anisotropic diffusion, contrast stretching, and high boost filtering. After pre-processing the images are segmented using the fuzzy C-means clustering segmentation. Features extraction is performed on the segmented images using the mean, skewness, kurtosis, gabour, grey level co-occurrence matrix, and discrete wavelet transform. The extracted images are merged with subject records using the canonical correlation analysis. The selected features are fed into the IDBNN to detect the condition of the osteoporosis prediction while the Dingo optimization algorithm by selecting the optimal weight. The proposed model has 94
To address numerous farmland challenges, Russia has been implementing abundant programs that have been demonstrated to significantly reduce the ecological degradation of farming. Food security and sustainable agricultural production constitute significant concerns for society. This research analyzes various data to explore these aspects in Russia, such as: soil moisture, net primary productivity (NPP), gross primary productivity, actual evapotranspiration (ETa), precipitation (mm), land surface temperature (°C), normalized difference vegetation Index, leaf area index, and digital elevation model (DEM), and crop grain production in 2010–2020. The goal of this study is to develop a linear regression model for predicting farmland gross primary productivity (GPP) by integrating remote sensing data with various environmental and agricultural parameters. The annual farmland GPP increased from 2010 to 2020. The results illustrate that the differences in crop cereal production data were not consistent with the annual pattern. Therefore, the recommended linear model for farmland GPP prediction performed significantly better in the regional studies. Our findings demonstrate that linear regression-based farmland GPP products can be utilized in the adequate assessment of agricultural ecosystems at the field and regional scales. It is also beneficial for food safety and security for the growing population.
Iris pattern recognition’s stability and uniqueness have significantly improved biometric authentication. The individual’s iris is the most promising biometric authentication method that can reliably identify an individual based on their unique characteristics. Traditional drawback of iris recognition systems is their vulnerability to such spoofing, which can compromise the effectiveness of biometric security, making it essential to enhance the detection of fraudulent attempts by the system. Iris recognition problems still exist even with sophisticated algorithms, especially in real-world settings where consumers might not be aware of any weaknesses. A deep learning-based segmentation and classification method for iris liveness detection utilizing quality metrics has been proposed in order to address this problem. Human eye images are first gathered and used as an input dataset. A Lightweight Attention Guided ConvNext Network (LACN) is used for image improvement and a Pixel Density based Trimmed Median Filter (PDTMF) to reduce noise. A Lightweight Dual Multiscale Residual Block-based Convolutional Neural Network (LDMRes-Net) is used for segmentation in order to recover the iris and pupil regions. These separated regions are used to calculate important iris and pupil region characteristics including as area, radius, and perimeter. Additionally, quality measures including pupil-iris ratio, iris-sclera contrast, iris-pupil contrast, grayscale utilization, and usable iris area are calculated. A deep learning classifier known as the Progressively Growing Adversarial Network with Dropout Layer (PGAN-DL) uses to determine iris liveness. Simulated findings show that the proposed technique achieves 98.1
An Erratum to this paper has been published: https://doi.org/10.3103/S1060992X26020013
Multi-agent pathfinding (MAPF) is a common abstraction of multi-robot trajectory planning problem, where multiple homogeneous robots simultaneously move in the shared environment. This paper addresses the heterogeneous MAPF problem, where a group of adaptive agents interacts with other agents (called impostors) that behave differently. The task remains cooperative, all agents should have the opportunity to reach their goals. We investigate how homogeneous methods can be enhanced for heterogeneous settings through three distinct approaches: planning-based, sampling-based, and learning-based methods. Our experimental framework employs the POGEMA benchmark to evaluate adaptive agents interacting with impostors following different policies (A* and PIBT). Our results demonstrate that all methods show significant performance improvements primarily with large agent populations, where frequent encounters with impostors necessitate conflict resolution. These findings indicate that while predictive modeling can enhance non-specialized algorithms when online training is impractical, learning-based methods offer superior adaptability to novel agent types in dynamic heterogeneous environments.
Wireless Body Area Networks (WBANs) are precisely defined as wireless networks comprising various sensors strategically positioned on the human body, these sensors have been either worn externally on the body or surgically implanted beneath the skin. Sensitive information is susceptible in many ways when it is transmitted across unsecure networks, therefore robust security measures are necessary to guard against possible attackers. Thus, the proposed model developed a secure and efficient patient monitoring using ElGamal-LCA for encryption with routing algorithm and MSDGCN based intrusion detection system. The process begins with a WBAN employing 12 sensors such as ECG, EMG, PPG, EEG, temperature, blood pressure, SPO2, respiration rate, accelerometer, glucose, gyroscope and galvanic skin response for capturing vital physiological signals from the human body. Then these readings are sent to a control unit which further aggregates the sensor data. For securing the data transmission Elgamal-Lightweight Cryptography Algorithm (Elgamal-LCA) is employed. Elgamal cryptosystem handles key generation while lightweight encryption encrypts the data. The data transmission causes interchannel interference due to overlapping signal from same or adjacent channels which are mitigated by utilizing a Stochastic Learning Algorithm (SLA) to prevent data loss and collisions. Once if interference is mitigated, data is transmitted to base station using Quality of service (QoS) based Minimal Latency Routing Strategy. At the base station intrusion detection was performed and the process involves preprocessing using Hyperbolic Tangent (HT) normalization and Slim Generative Adversarial Imputation Network (SGAIN) for imputing missing data followed by classification utilizing Modified Spatial Dynamic Graph Convolutional Network (MSDGCN) with Dynamic Composable Multi-Head Attention (DCMHA) for effective detection. Finally, alerts are sent if an intrusion is found otherwise data is stored securely in the cloud. The proposed approach achieves an execution time of 77.51 sec, packet loss of 4.7
In human-computer interaction, Facial Emotion Recognition (FER) is essential, particularly in fields like behavioral analysis and psychological therapy. Perceiving emotions accurately from facial expressions can enhance communication and interaction between humans and machines. However, the wide range of human faces and image variations, including different lighting conditions and facial poses, makes it difficult to achieve accurate and robust FER using computer models. In this work, we chose to work exclusively on the FER2013 dataset to address its complications and complexities in terms of feature extraction. To analyze its impact, we employed a Deep Convolutional Neural Network (DCNN) and pre-trained models that have been adjusted specifically for emotion recognition. This work takes into account a pre-processing step that concentrates on image resolution, histogram equalization, and data augmentation. We achieved a higher accuracy of 76
The problem of modeling the movement of autopiloted vehicles in a mixed traffic stream of autopilots of various manufacturers, in which there are no collisions, is solved. A neural network model is proposed that implements reinforcement learning for an unmanned vehicle model integrated into a microscopic model of a mixed traffic flow. Computational experiments are being conducted with the proposed model. It was hypothesized that trained agents would work worse together than with the original automated objects in which they were trained. In the first experiment, for a closed multi-corridor circle, the percentage of implementation of trained agents gradually increased. In the second experiment, different trained agents were trained on a single-lane circle and then launched together. In the third experiment, agents are trained in different environments and run together on a track with multiple corridors. As a result of the experiments, the hypothesis was not confirmed. When using different learning environments, agents trained in these environments interact more effectively with each other in a common system than with other types of agents.
Biometric authentication powered by Artificial Intelligence (AI) has arisen as a vital solution for ensuring secure access to digital healthcare data. By leveraging advanced AI-driven algorithms, such systems can accurately recognize and verify users based on unique biological traits. However, various existing authentication method suffer from limitations such as noise distortion, poor illumination handling, redundant features, weak multimodal fusion and reduced capability in distinguishing between genuine and fake biometric inputs. To address these challenges, physics inspired deep learning based two-step biometric authentication verification framework is developed to enhance healthcare data protection. The system integrates originality verification of iris and fingerprint modalities to achieve high reliability and precision in healthcare user verification. Initially, biometric images like fingerprint and iris are pre-processed using Wavelet-Inspired Invertible Network (WINNet) for Denoising and the Detail-Enhanced Attention Network (DEA-Net) for illumination and contrast enhancement. Textural features are then extracted using the Hexadecimal Local Adaptive Binary Pattern (HLABP) technique. Both finger print and iris features are adaptively combined through the Adaptive Feature Fusion Mechanism (AFFM) to minimize redundancy and improve representational strength. Finally, the Physics-Informed Neural Network based First-Order Reliability Method (PINN-FORM) classifiers performs biometric recognition to differentiate real from fake data. Upon successful verification, a secure QR code is generated through the Stylize aEsthEtic (SEE) mechanism exclusively for legitimate users. The users then scan the QR code for further biometric verification. If the biometric matches, the user can access the data, otherwise the request is declined. The suggested multimodal biometric authentication framework demonstrates exceptional performance, achieving a precision rate of 97.87
The performance of picking robots for fruit object detection is crucial in agricultural environments. However, most existing detection models struggle to perform well in agricultural settings due to problems in detection accuracy, computational resource consumption, and real-time processing. To address these challenges, we propose a navel orange detection model called CWG-YOLOv8 based on YOLOv8, which can achieve accurate detection of navel oranges in agricultural environments. Firstly, we introduce a Convolutional Block Attention Module (CBAM) to enhance the backbone network and improve the generalization ability of the model. Secondly, Wise-IoU (WIoU) v3 as the bounding box regression loss function is employed, and a wise gradient allocation strategy is incorporated to emphasize high-quality samples, thus enhancing the model’s localization capability. Finally, we design a lightweight GhostNet module that effectively integrates shallow and deep features to reduce computational cost and speed up detection. The experimental results show that the number of parameters and the number of floating-point operations (FLOPs) of our model are reduced by 42.35 and 36.59
Colon cancer is a type of cancer that affects the colon (large intestine) or rectum. It is one of the most common kinds of cancer worldwide and can cause severe harm and death. Early detection of colon cancer is especially difficult because of cancer cells overlap, making identification more difficult. Also, classifying cancerous cells in histopathology images is difficult due to the complex inter-class and intra-class dependencies. It can be challenging to distinguish between normal and cancerous cells because the underlying tissue structures often merged and have similar morphological structures. This convolution of structural features contributes additional complexities that hinder accurate evaluation and identification. To address these drawbacks proposed an Artificial Recurrent Neural Network with Levenberg-Marquardt Method (ARNN-LMM) based elapid encryption to improve security and predict colon diseases using IoT-enabled devices. Initially, Colon Cancer Histopathological Images are collected to serve as the input image. In order to reduce the disturbances in the background, the pre-processing of the raw images is done first using a pixel-wise thresholding (PWT) method, which is used to provide the image with a better appearance in addition to reduction of the noise. A wavelet domain transformer (WavEnhancer) is then used to enhance clarity at the pixel level and hence effectively enhance the overall image quality. Circular Mesh Network (CirMNet) is a shape based feature extraction technique, which extracts structural, statistical, and property-based features of image. The refined features are fed into a classification employing ARNN-LMM to detect colon abnormality. The trained net model is encrypted using elapid encryption and stored in cloud for ensuring secure access. In Disease Prediction Phase, an Internet of Things (IoT) device captures patient data and transmitted to the cloud, where the model is decrypted and analyze the features to predict the patient has disease or non-disease by a trained model. An suggested model attains 97.45
Image forgery detection is the process of identifying manipulated or altered images to determine their authenticity, ensuring the integrity of digital media. Image forgery was predicted to maintain trust in digital media, prevent misinformation, and ensure the credibility of visual evidence in various fields like journalism and law enforcement. Traditional methods for predicting image forgery relied on manual inspection and basic image processing methods, those techniques were time-consuming, required expert knowledge, and were prone to human error, making them less reliable and efficient. Recently presented Artificial Intelligence (AI) techniques, automate the detection process by learning from large datasets, offering increased accuracy and faster processing times. However, such methods need large amounts of labelled data and a good amount of computational power to train the models properly. To address these limitations, proposed Global Convolutional Context Networks (GCCNet) were employed to identify a fake images. The image is first taken from a dataset that was used to detect copy moves. The Quantum Wavelet Transform Filter (QWTF) and RetiNex algorithm are used to pre-process the input image. That is the original image’s noise is eliminated using QWTF, and the pixel contrast is improved using RetiNex. The copied portion of the pre-processed images is then segmented using the Pyramid Scene Parsing Network (CAP-PSPNet), which is based on Contour Aware Processing to make the segmentation edges sharper. The image’s segmented portion is then passed to a hybrid Attention-based Shuffle-Net V2 (ATSNetV2) with Global Convolutional Context Networks (GCCNet) to predict real or fake images. As a result, the model attained an accuracy of 96.60
Wireless Sensor Networks (WSNs) consist of numerous small, multifunctional sensor nodes deployed in target areas to collect and transmit environmental data. Due to the energy constraints of battery-powered nodes, balancing energy consumption with network performance remains a significant challenge. To address this, an approach integrating Artificial Neural Network (ANN)-based outlier detection with a secure and energy-efficient data transmission mechanism using an enhanced LEACH protocol is proposed. Nodes are initialized in the sensing region to gather informations. Then Artificial Neural Network is utilized to identify the outliers from the deployed nodes. Then, cluster formation and cluster head selection are done by employing a stable and secure LEACH protocol. A hybrid reputation-based secure data transmission (HRSDT) is employed to transfer data from its base station after CH selection. Community and QoS reputation rates are employed for sensor node reputation calculation. Among all nodes, the node with the highest HRSDT is regarded as the forward node for safe data transfer. According to the assessment results, the proposed approach obtains 92
