Journal of Intelligent Systems and Internet of Things

In the area of digital information, establishing the authenticity of an image has grown to have greater significance as more and more persons have access to sophisticated image editing technologies. There is however a challenge in detecting such a forgery since it is usually very realistic and it is hard to know the difference between the real images and the fake ones. This paper aims at creation of a mechanism of identifying forged images based on Multiple Image Splicing Dataset (MISD) as a reference point. The suggested system will help to improve the results of the forgery detection, paying particular attention to the images processing during some of the pre-processing steps Firstly, converting colors into the hue-based histograms and RGB histograms, and hue-based histograms in an HSV, in comparison between the original and forged image, its HSV histogram, and its grayscale histogram, etc. Lastly, compute MSE and SSIM original and forged image. The implementation results showed that average value of MSE and SSIM metrics on Multiple Image Splicing Dataset (MISD) equal to 184.82 and 0.65 respectively that means the suggested method proved the efficiency of the technique to identify forged images as quickly as possible but still retain accuracy.

Gesture recognition serves as a key enabler for natural and intuitive human–robot interaction (HRI) in smart automation and assistive systems. However, achieving real-time performance with high recognition accuracy remains a significant challenge due to dynamic background variations, occlusion, and complex spatio-temporal dependencies in gesture sequences. This paper presents a real-time attention-based CNN-RNN framework for robust gesture recognition and adaptive HRI in dynamic environments. The proposed system utilizes Convolutional Neural Networks (CNNs) for spatial feature extraction from sequential video frames and Bidirectional Recurrent Neural Networks (BiRNNs)—integrated with an attention mechanism—for modeling temporal dependencies and focusing on discriminative motion cues. The attention layer enhances interpretability by prioritizing salient gestures and reducing background noise. A hybrid optimization strategy, combining adaptive learning rate scheduling and regularized dropout, ensures computational stability and generalization across gesture datasets. Experiments conducted on benchmark datasets such as NVIDIA Dynamic Gesture (NvGesture) and ChaLearn IsoGD demonstrate superior performance, achieving an accuracy of 97.8% and a real-time inference speed of 34 FPS, outperforming baseline CNN, 3D-CNN, and LSTM architectures. The proposed framework effectively balances accuracy, latency, and interpretability, making it suitable for real-world HRI applications, including service robotics, industrial automation, and assistive technologies.

Social media platforms have become pivotal arenas for the public to express emotions, opinions, and sentiments. While traditional sentiment analysis methods predominantly focus on textual data, they often overlook the rich emotional context embedded in images shared alongside posts. This paper presents a novel framework that integrates Visual Sentiment Analysis (VSA) with Natural Language Processing (NLP) techniques to enhance the understanding of public sentiment in social media content. By leveraging deep learning-based feature extraction from images (using pre-trained CNN models) and combining them with transformer-based text analysis (such as BERT), the proposed multimodal sentiment analysis model captures nuanced emotional expressions more effectively than unimodal approaches. Experiments conducted on benchmark datasets, including Twitter and Instagram posts, demonstrate a significant improvement in sentiment classification accuracy and contextual interpretation. The study highlights the potential of integrated sentiment analysis systems in applications such as brand monitoring, political opinion tracking, and mental health detection.

The exponential growth of the Internet of Things (IoT) ecosystem has amplified concerns regarding data privacy, trust management, and cyber resilience in decentralized environments. Traditional perimeter-based security models are inadequate for heterogeneous IoT networks that operate across multiple domains. To address these challenges, this paper proposes a Blockchain-Augmented Zero Trust Architecture (BZTA) integrated with a hybrid intrusion detection mechanism for achieving secure, verifiable, and adaptive threat mitigation in decentralized IoT frameworks. The proposed BZTA employs blockchain-based identity verification to ensure device authenticity and policy-driven Zero Trust enforcement to validate every access request dynamically. A federated intrusion detection model built using Long Short-Term Memory (LSTM) and Graph Attention Networks (GAT) identifies anomalous communication patterns, while smart contracts facilitate tamper-proof logging and automated response coordination. The integration of Proof-of-Trust (PoT) consensus enhances scalability by minimizing latency during transaction validation. Experimental evaluations conducted on simulated IoT network datasets demonstrate a detection accuracy of 98.6%, false positive rate of 1.8%, and an average latency reduction of 22% compared to traditional IDS and standalone blockchain systems. The proposed BZTA framework effectively balances security, scalability, and interoperability, providing a resilient foundation for next-generation decentralized IoT infrastructures.

This research presents a comprehensive analysis of path planning and optimization techniques in mobile robotics, focusing on both classical algorithms and modern intelligent approaches. The study systematically reviews fundamental methods such as Dijkstra’s algorithm, the A* search algorithm, and artificial potential fields, together with evolutionary optimization approaches including genetic algorithms and swarm intelligence. It also explores the application of machine learning and deep reinforcement learning models that allow robots to adapt dynamically to complex and changing environments. The comparative evaluation highlights the strengths, weaknesses, and suitable application areas of each approach across scenarios involving obstacle avoidance, energy efficiency, real time adaptability, and multi robot coordination. Particular attention is given to the challenges of uncertain and dynamic environments, computational scalability, and sensor noise, which continue to limit the performance of autonomous navigation systems. By consolidating current advancements and emerging trends, this study provides a structured overview and critical synthesis of existing methodologies, offering a valuable reference for researchers, engineers, and practitioners. It also identifies important research gaps in intelligent hybrid planning, context aware learning and energy constrained mobility, outlining promising directions for the future development of autonomous robotic navigation systems.

Rapid spread of Corona virus 2019 (COVID-19) is predictable to create high contact on healthcare organization. Early detection of this disease is required to make precise treatment that further helps to increase the survival rate of humans. However, detecting the COVID-19 at beginning stage is one of a major challenge in the world because of rapid disease spread. Various existing methods are developed to detect the disease, but generating accurate result at the beginning stage still poses a complex task in the medical research community. Hence, an effective mechanism is modeled in this research to predict the pandemic at early with the time-series data using proposed Water Poor and Rich optimization-based Deep Recurrent Neural network (WPRO-based Deep RNN). Accordingly, proposed method is highly effective in generating the most appropriate results through deep learning classifier based on the high dimension features. However, the high dimensional data is generated through the data augmentation process by employing oversampling technique. The proposed method is more robust and increases the efficiency of the optimization algorithm by attaining global convergence results based on the fitness measure. Accordingly, the technical features of time series data to improve effectiveness of developed model. However, the proposed WPRO-based Deep RNN produced minimum Root Mean Square Error (RMSE) as well as MSE values of 0.4 and 0.1714 for confirmed cases, and obtained lesser MSE and RMSE values of 0.1887 and 0.433 for the cases of death. Moreover, proposed model achieved minimal RMSE and MSE of 0.447 and 0.1901 for the recovered cases.

This paper introduces an advanced method for face recognition in video surveillance systems, leveraging only a single reference image per individual. The challenge of recognizing faces in video is addressed, considering issues like pose variations, occlusions, and lighting changes. The proposed approach utilizes 3D Morphable Models (3DMM) to generate a 3D face mesh from the reference image, facilitating robust face alignment and recognition across video frames. A Convolutional Neural Network based pipeline is employed for face detection, pose estimation, and extraction of invariant features, while an optimization framework refines landmark positions and depth maps for accurate 3D reconstruction. The system performs exceptionally well on the CASIA-WebFace Dataset, with 97.00% pAUC (20%) in surveillance mode and 98.69% in identification mode for frontal views. With an efficiency of 16.72 FPS on modest hardware, the system proves its practicality for real-world deployment. The method incorporates synthetic data augmentation and Random Subspace Methods to enhance adaptability to domain-specific conditions. Compared to existing methods like Eoe-SVM and CCM-CNN, the proposed system demonstrates a superior balance between accuracy and computational efficiency, particularly in Single Sample Per Person (SSPP) scenarios. By focusing on single-reference image recognition, the system offers a promising solution for large-scale surveillance applications, where video footage typically contains multiple poses, expressions, and lighting variations. The results highlight the system's effectiveness and efficiency, making it an excellent alternative for real-time face recognition in complex and dynamic surveillance environments.

The Internet of Things (IoT) has greatly changed many aspects of human life and is now a vast distributed systems network of interconnected devices that have embedded sensors; however, the battery life of these sensor nodes is limited and requires constant maintenance. Furthermore, IoT networks operating as distributed systems are vulnerable to security threats, like de-authentication and Disassociation Denial-of-Service attacks, which exploit vulnerabilities in Wi-Fi devices. While artificial intelligence, including machine learning, has been integrated into intrusion detection systems to enhance detection of cyberattacks, there is an increasing need for improved accuracy, scalability, efficiency, and IoT-specific security solutions. This paper proposed a novel model, Hybrid Optimization-based Clustering with CNN-Based De-Authentication (HOCCNN), designed to concurrently address both energy conservation and security issues in IoT-enabled heterogeneous wireless sensor networks (WSNs). The HOCCNN adopts a hierarchical clustering technique optimized using the bio-inspired Osprey Optimization Algorithm (OOA) for dynamic and energy-efficient Cluster Head (CH) selection. Additionally, we introduce a CNN model to detect and mitigate De-authentication attacks in HOCCNN by utilizing deep learning techniques and provide a more accurate threat detection solution even in the resource-constrained environment. The performance of HOCCNN was evaluated using MATLAB against existing baseline methods in terms of parameters like packet delivery ratio, network throughput, network lifetime, end-to-end delay, average energy consumption, data accuracy, and data overhead. The model demonstrates superiority over state-of-the-art baselines. Results show significant improvements. 99.1% accuracy in attack detection, 54.18% energy consumption, 6.76 s network lifetime, 0.985 packet delivery ratio, and 53.198 Mb/s throughput. These results prove that HOCCNN is a complete design to achieve scalable, secure, and energy-sustainable HWSNs in IoT.

The transition from conventional vehicles to electric vehicles (EV) represents an important development in the field of sustainable transportation. To prevent concerns about battery drain, the use of EVs requires the establishment of sufficient charging stations (CS) to recharge vehicle batteries. In Iraq, the infrastructure of electric vehicle charging stations (EVCS) is still limited, which reduces the reliance and reliability of EVs. This study assessed the economic efficiency and feasibility of optimizing hybrid renewable energy systems (HRES) for EVCS in three cities of Iraq addressing the growing demand for renewable energy due to concerns regarding fossil fuel depletion, environmental sustainability, and escalating conventional energy expenses. Hybrid Optimization Model for Multiple Energy Resources (HOMER) program was used considering weather data, load profiles, and equipment specifications. The results indicated that the system with a capacity of 300 kW of photovoltaic (PV), 100 kW of generator (GEN), and 78 units of batteries is found to be the optimal system in all three cities, with the lowest cost of energy (COE) around 0.025 $/kw. The renewable energy fractions of the optimal system in Mosul, Baghdad, and Basrah are 53%, 52.7%, and 52.7%, respectively. This setup achieves annual energy production of 704351 kWh from PV and 509681 kWh from GEN. This arrangement keeps the battery storage at a high state of charge (SoC), guaranteeing system stability and prolonging the battery's life. The system's capacity to reliably fulfil load requirements with less dependence on the DG. These results provide valuable insights into the deployment of HRES to achieve a more sustainable environment.

Software-Defined Networks (SDNs) offer unparalleled network control flexibility, yet efficient load balancing is still challenging in dynamic environments. DeepBalance is a novel framework presented in this paper, which deploys dynamic load balancing in SDNs using Deep Reinforcement Learning (DRL). Our solution employs a Deep Q-Network (DQN) agent, which learns the optimal routing policies by monitoring network states and being rewarded based on load distribution. DeepBalance continuously tracks link utilization and intelligently reshifts traffic to alleviate congestion and achieve maximal throughput. We employ a comprehensive simulation environment, which emulates actual network conditions and traffic patterns. Experimental results demonstrate that DeepBalance significantly outperforms traditional load balancing techniques, lowering link utilisation variance by 37% and total throughput by 28% over shortest-path routing. The infrastructure adapts with changing traffic patterns automatically without the necessity of manual reconfiguration, thus naturally circumventing hotspots by making forward-looking path decisions. Additionally, our visualizations illustrate how the DRL agent learns over time to distribute network load more evenly across alternative paths. DeepBalance is a strong candidate for autonomous network optimization in future SDN deployments.

More environmentally friendly standards may be implemented as work environments, lifestyles, and our conception of a fulfilling life evolve. The COVID-19 pandemic highlighted the need for adaptable systems and revealed the flaws in our routines. Because smart cities are more flexible than traditional urban areas, they are becoming more and more important. While supporting citizens is the main goal of these networked smart city components, they also unintentionally enhance urban environments. This paper uses a methodical approach to investigate smart cities, breaking down and analyzing each component to clarify their beneficial interactions. This paper provides a direction for future research through its discussion of problems, challenges, and barriers related to the urban environment that affect the development of smart cities. Real-time monitoring is made possible by connecting these devices to the internet. The spacing between lighting poles significantly influences the overall uniformity and illuminance. This paper describes the architecture of Internet of Things (IoT)-based smart public smart utility system using forecasting techniques that interconnected with the sensors using IoT stack. Sensors are made to gather the timely data from different utility applications such as lighting, CCTV cameras, water usage, wastage volume, etc. The paper demonstrates the potential synergies between IoT and artificial intelligent for supporting smart cities. We deployed three convolutional neural networks namely: AquaNet, PredWasting and LightSage for forecasting the water requirements, wasting volume and light consumption in smart cities. Results shown that PredWasting is outperformed with 99.21% of accuracy over the other models.

Due to the Internet's growing importance in our lives, Software-Defined Networking (SDN) networks have experienced high load traffic issues. Thus, network load has increased, lowering quality of service (Qos) performance. Modern networked systems depend on communication channels to transmit data between sources and destinations.  High traffic loads exacerbate packet distribution inefficiencies, causing network congestion in specific channels, compromising these communication channels. Congestion delays packet delivery and generates significant packet loss, reducing network dependability and efficiency. Communication channels' improper packet allocation along accessible paths is the fundamental issue. Some paths are overcrowded during peak traffic, while others are underused.   Bottlenecks slow packet transit and increase packet loss due to this imbalance. Current packet distribution techniques don't adapt effectively to dynamic traffic, resulting in poor network performance. Current traffic management solutions often rely on load balancing algorithms, but these methods may not adequately account for the dynamic and unpredictable nature of high-load traffic. This paper introduces Adaptive Load Balancing using Reinforcement Learning (ALBRL), which uses Q-learning and deep reinforcement learning to distribute traffic in real time in SDNs with high traffic loads. This model uses more network-specific indicators including packet loss ratio, latency, Jitter, and traffic pattern history to improve decision-making. ALBRL outperformed static routing and Q-learning with 15.34(ms) average delay, 2.11(ms) jitter, and 7.89% packet loss ratio.

Skin image segmentation serves as a vital undertaking in medical image analysis, specifically in dermatology, since it enables the detection of skin diseases and the assessment of effectiveness of treatments. Segmenting skin lesions from photographs is a crucial step in working towards this patchive. Nevertheless, the work of segmenting skin lesions is difficult due to the existence of both artificial and natural deviations, inherent characteristics like the shape of the lesion), and deviations in the circumstances during which the images are obtained. In recent years, researchers have been investigating the feasibility of utilizing deep-learning models for skin lesion segmentation. Deep learning methodologies have exhibited encouraging outcomes in various image segmentation initiatives, thereby preventing the possibility of automating and enhancing the precision of skin segmentation. This paper introduces a new hybrid method, named the CBi-BERT framework, aimed to improve the results and architectures of medical image segmentation or patch detection tasks. This architecture employs Convolutional Neural Networks (CNNs) for feature extraction as well Bidirectional LSTM-based encoders to process sequence information and BERT based attention collection across the strongest features. Image normalization, resizing and data augmentation techniques are applied in the proposed method to deal with major challenges faced during medical imaging such as rate of image quality differentiation from noise or bias between benign vs. malign features. We evaluate the performance of CBi-BERT to those benchmark datasets and state-of-the-art models, showing that CBi-BERT outperforms them in all relevant metrics such as Intersection over Union (IoU), recall, mean average precision (bin-MAP) DICE coefficient. Specifically, for images sized 256x256 the model achieved IoU =0.9, recall=0.92, mAP=0.89 and Dice coefficient: =0.91 thereby outperforming some advanced state-of-the-art models as ResNet50, VGG16, UNet, EfficientNet-B-01 Our results show that the framework is able to detect and segment important structures in medical images with high precision which makes it a compelling tool for AI based Healthcare solutions.

The implementation of Intrusion Detection Systems (IDS) remains crucial for network security yet high-dimensional data alongside class imbalance issues decrease their functionality. Machine learning-based IDS models, which use traditional approaches experience difficulties in providing explanations about their prediction results. An IDS framework enhancement with explainable AI (XAI) methods aims at improving the system's transparency throughout this study. The data processing includes KNN imputation combined with K-Means SMOTE to handle missing information and class imbalance problems. When selecting features the model uses a merged methodology combining Pearson Correlation with Mutual Information and Sequential Forward Floating Selection (SFFS) algorithms for optimization. Light Gradient Boosting Model (LGBM) serves as the classification model that produces higher accuracy than competing methods with 90.71% for UNSW-NB 15 and 96.98% for CICIDS-2017. By using SHAP-based explain ability, the system provides worldwide and specific model interpretations that enable users to trust IDS prediction results. The experimental findings validate that the proposed methodology succeeds in simplifying the system while improving its classification functionality and delivering stronger interpretability properties to tackle weaknesses of current IDS technologies. The examination presents important findings for the development of secure network protection technologies, which operate with transparency.

Internet of Vehicles (IoV) is the later application of VANET and is the fusion of the Internet and IoT. With the advancement in innovation, individuals are investigating a traffic environment wherever they would have the extreme cooperation with their environment including other vehicles. The Internet of Vehicles (IoV) was created so that vehicles can communicate with each other in an infrastructure environment. The prerequisite is to form a more secure trip in an IoV environment with the least delay and high packet delivery rate. This guarantees that all information is received with negligible delay to maintain a strategic distance from any mishap. This paper presents a new position-based routing algorithm called Position-Based Connectivity Aware Routing (PBCAR) for IoV that covers sparse and coarse regions of vehicles. It takes advantage of the Internet and street format to progress the execution of routing in IoV. The PBCAR algorithm uses a GPS real-time chasing system to find traffic information for forming position-based paths from the source node to the destination node. The PBCAR algorithm has been simulated using SUMO and Network Simulator and compared with AODV and GPSR. The results show that the PBCAR algorithm obtains exceptional results considering the several simulation parameters.