Al-Farabi Journal of Engineering Sciences

Arabic handwriting recognition represents one of the most demanding problems in pattern recognition due to the distinct structural and linguistic properties of the Arabic script. Unlike Latin-based writing systems, Arabic text is inherently cursive, displays position-dependent character shapes, and relies heavily on dots and diacritics to convey meaning. These intrinsic characteristics contribute to segmentation ambiguity, shape similarities, and high variability in handwriting styles, making traditional OCR solutions inadequate. This research examines the conceptual foundations and technological evolution of Arabic handwriting recognition, beginning with handcrafted feature engineering—including structural, geometric, and global transformation descriptors—and progressing to the current dominance of deep learning paradigms. A detailed architectural survey of state-of-the-art deep learning models is presented, highlighting the role of convolutional neural networks (CNNs), recurrent architectures such as BiLSTM, attention mechanisms, transformers, and segmentation-free classification via Connectionist Temporal Classification (CTC). Additionally, a complete deep learning framework is developed and illustrated, including model design, dataset structure, and decoding methodology, supported by example code for practical implementation. The experimental context further demonstrates how deep learning enables automated high-level representation learning and robust recognition without manual segmentation or handcrafted features. Collectively, the findings reinforce that the shift toward end-to-end deep learning architectures—especially those integrating visual, sequential, and contextual reasoning—provides a viable solution to the long-standing challenges posed by Arabic handwritten text. This research contributes to both theoretical understanding and applied methodology, offering a comprehensive foundation for future advancements in intelligent document processing, digital archiving, and Arabic language technologies.
Chapter One

Artificial intelligence (AI) has rapidly evolved from a technical innovation into a strategic driver of organizational transformation. While prior research has examined its potential, this study advances the concept of learning processes, with a particular focus on workflows that continuously evolve through the integration of AI capabilities. Drawing on recent academic literature and secondary case evidence from leading firms such as Google, IBM, Tesla, Amazon, and Netflix, the paper analyzes how AI is reshaping business processes across multiple sectors. The findings highlight tangible benefits including efficiency gains, cost reductions, improved decision-making, and innovation in business models. At the same time, significant technical, organizational, and ethical challenges emerge, ranging from data governance and system integration to workforce adaptation and algorithmic accountability. Looking forward, the study identifies five trends: explainable AI, convergence with emerging technologies, advanced cognitive automation, democratization of AI, and ethics-driven governance, that will define the next phase of intelligent business. The contribution of this work lies in framing AI not as an isolated tool but as a catalyst for adaptive, learning processes aligned with strategic and cultural objectives. The paper provides both scholars and practitioners with a structured framework to understand, implement, and govern AI for sustainable competitive advantage.

Wireless Sensor Networks (WSNs) have strict energy resource problems that affect their lifetime in IoT environments. This work proposes E-LEACH, an improved version of Low Energy Adaptive Clustering Hierarchy that alleviates these issues by incorporating two effective strategies: a new threshold function with a denominator square for decreasing the rate of rotation for cluster heads, and Euclidean-distance clustering for improving communication across clusters. Network simulation experiments using MATLAB with different network densities (100-500 sensors) in 1000x1000 m^2 topologies showed significant improvements. Compared with standard LEACH, E-LEACH showed 88% improvement in network lifetime extension, 71.86% improvement in energy consumption efficiency, and 49.5% improvement in throughput performance. Packet Delivery Ratio was improved from 0.00955 to 0.01001, while average delay was reduced from 2.45 to 2.01 seconds.

AA7075-T651 aluminum alloy is indispensable for weight-critical aerospace and automotive structures due to its exceptional strength-to-weight ratio, yet its weldability remains challenging via conventional fusion techniques. Friction stir welding (FSW) with a tungsten carbide (WC) tool offers a solid-state alternative that minimizes defects but induces microstructural degradation—particularly dissolution of strengthening precipitates and grain boundary alterations—severely reducing joint toughness, especially in the nugget zone (NZ), this work implemented post-weld heat treatment (PWHT) to restore precipitate distribution, systematically evaluating T6 (peak aging) and T7 (overaging) cycles on AA7075 FSW joints, welding employed a WC tool (18 mm shoulder, 7.2 mm pin) at 800 RPM/90 mm/min. PWHT included solution treatment (480°C/1h) followed by T6 (120°C/24h) or T7 (105°C/8h + 170°C/16h) aging. Mechanical and microstructural characterization included Vickers hardness mapping, tensile testing (ASTM E8), Charpy V-notch impact tests (ASTM E23) with NZ-notched specimens, and OM/SEM fractography, results indicated that T6 restored 94% of base metal tensile strength (538 MPa) but achieved only modest toughness (12.1 J), t7, however, tripled NZ impact energy (21.5 J) versus the as-welded state (7.2 J) by coarsening intragranular η-phase precipitates (115 nm), reducing grain boundary precipitate density by 52% (1.2 μm⁻¹), and narrowing PFZs to 45 nm. Fractography confirmed fully ductile fracture in T7 specimens, contrasting with brittle intergranular failure in T6, the T7 cycle is identified as optimal for NZ toughness enhancement, providing manufacturers a pathway to improve impact resistance in safety-critical FSW AA7000 applications.

Abstract: Metal cutting is a fundamental industrial process, yet its complexity involving high strain, heating, and diverse operating conditions makes optimization challenging. This study focuses on optimizing two critical parameters in machining: time and energy consumption. Employing the Finite Element Method (FEM) via ANSYS software, the research investigates the intricate mechanics of machining processes. The methodology involved detailed modeling of the problem’s geometry, mesh discretization, and convergence analysis to simulate an end mill machining operation. Key findings reveal that tool-workpiece contact points consistently experience maximum stress. Furthermore, an optimal rotational speed between 4000 and 6000 rpm significantly reduces momentum fluctuations and shocks. While kinetic energy increases with rotational speed, reaching a plateau, increasing speeds beyond 8000 rpm yield diminishing returns in energy fluctuation reduction. Importantly, a growing disparity between work done and total energy exchanged at higher speeds indicates increased energy loss. This research provides valuable insights into enhancing machining efficiency and reducing energy waste through FEM analysis, offering practical implications for industrial applications.”

Climate change represents one of the most significant challenges of our time, affecting various ecological and health systems across the globe. As temperatures rise and weather patterns shift, the dynamics of bacterial pathogens are also changing. These changes can have extreme implications for public health, agriculture and biodiversity. Affecting human health, Human pathogenic bacteria resistance (HPBR) infections mediated by the environment are considered a substantial cause of global health losses. However, the biogeography of HPB and their response to climate change remain largely unknown. HPB are widely present in the global environment, and their distribution follows a latitudinal diversity gradient. Climate and anthropogenic factors are identified as major drivers of the global distribution of HPBR. Our predictions indicated that by the end of this century, the richness, abundance, and invasion risk of HPBR will increase globally, with this upward trend becoming more pronounced as development sustainability declines. Therefore, the threat of environmentally mediated HPBR infections to human health may be more severe in a world where anthropogenic activities are intensifying and the global climate is warming. The objective of the study has comprehensive climatic and biological frameworks. This strategy has effectively identified relevant international research information and highlighted areas where further study is needed. Climate change is likely to affect key aspects of bacterial epidemics in various ways, often exacerbating outbreaks but sometimes having a mitigation impact depending on the pathosystem and other factors.

This paper contributes to the advancement of blockchain technology in the financial sector and sets a new standard for payroll management in the digital identities, proposed model involves: data generation, encryption, blockchain creation, and summary reporting. Create fake datasets by bootstrap and create good-looking XML documents and datasets has been created for 25000 records. Each record has (‘id', 'name', 'sex', 'birthdate', 'certificate', 'title', 'department', 'level', and 'salary'). Read datasets in the model and generate unique key generator has been calculated by lower key size, upper key size, digits, and punctuation. Generate random characters for each category without duplicates, combine all categories into a single list, shuffle the list to randomize the characters, and then convert the shuffled list back to a string and return it as the AES key. The model has been tested by National Institute of Standards and Technology (NIST). Results show that the model working perfectly, therefore show the model was successfully passed with 12 features and fall with 3 features, mean that 80% passed 20% fall, thus test has been done under key encryption test. Otherwise, model has been tested on cipher text, and the result show that the model was successfully passed with 11 features and fall with 4 features, the percentage of successfully was 74% and fall in 26% in Figures (5.1) and (5.2) shows the test of model.

The prevalence of Industrial Internet of Things (IIoT) systems has significantly advanced automation processes in industrial environments, simultaneously exposing critical infrastructure to high-level cyber-attacks. Recent attacks, such as zero-day and multi-stage attacks, are particularly challenging due to their dynamic nature and lack of previous attack signatures. Traditional industrial intrusion detection systems (IIDS) mostly rely on point-based or fixed-time interval anomaly detection, which limits their ability to predict temporal attack evolution and provide early alerts. This work introduces an unsupervised, explainable temporal inference framework that models cyber-attacks based on behavior evolution, rather than categorizing them. Unlabeled raw data is directly inferred into normal system dynamics using a deep temporal model. Deviations are construed as anomalies in behavior. These abnormalities are then narrowed down to capture long-term, system-wide changes while filtering out short-term operational changes. A probabilistic time model of inference is then implemented to infer latent states and transitions in the system. This allows us to discern and interpret multi-stage attack progression without labeling attack values. The proposed model is tested on popular industrial and IIoT benchmark datasets such as SWaT, WADI, and ToN-IoT in the presence of the most severe zero-day conditions. Empirical findings support detection rates of 98.4%, 97.2%, and 96.9% on SWaT, WADI, and ToN-IoT, respectively, with false alarms at less than 2.5%. Additionally, the framework achieves close-to-real-time detection with an average detection delay in the sub-second range.

Abstract
The growing demand for low-carbon construction materials and effective utilization of industrial by-products has intensified research interest in geopolymer concrete as a sustainable alternative to ordinary Portland cement. Among various precursors, fly ash–based geopolymer systems have demonstrated significant potential; however, their performance is strongly influenced by the variability in fly ash availability, chemical composition, and reactivity. To overcome these limitations, the incorporation of supplementary aluminosilicate sources has become increasingly important. In this context, waste glass powder, characterized by its high amorphous silica content, has emerged as a promising modifier that not only enhances geopolymer performance but also contributes to sustainable solid waste management. This review critically examines recent advances in fly ash-based geopolymer concrete modified with waste glass powder, with particular emphasis on geopolymerization mechanisms, material reactivity, and performance enhancement strategies. Findings from the literature indicate that partial replacement of fly ash with waste glass powder commonly in the range of 20 to 30% promotes geopolymer gel development, refines pore structure, and enhances

Abstract
Fashion, digital media, and design are some of the creative industries in business that face great challenges in making marketing decisions due to competition and the dynamics of the environment. Association Rule Mining (ARM) has proven to be very useful in discovering hidden patterns and relationships in consumer behavior data used in advertising and providing suggestions. This is because of the lack of an ability to manage, store, and share the discovered rules. This project seeks to link knowledge management (KM) and ARM to create a marketing decision model that is adaptable and sustainable. This link ensures the storage and management of the extracted rules in a single knowledge management repository, making it inevitable to keep marketing decisions current and aligned to changing consumer behavior trends. The report also reveals a critical gap in previous studies in the literature review section, indicating that although many studies have utilized ARM and deep learning to process and analyze different datasets (retail, fashion, Twitter, e-commerce, and live streaming), they have been limited in their sustainability and reusability due to the lack of knowledge management systems. This study fills this gap by linking KM and ARM to make things more accurate and adaptable in business environments characterized by constant changes.

This work presents a method for image encryption depended on the chaotic map with AES-Substitution-box that is one of a most essentials and just non-linear attribute of block cipher. Encryption algorithm consists of the encryption keys generation by using the proposed algorithm with its permutation and diffusion. This scheme is utilized to encrypt the images (color) with any size by using (substitution-permutation) table. The construction of Logistic map with one way-S-box is to use the generated key sequence, it is more robust to algebraic attacks than chaotic. One-way S-box utilized by adopting appropriate affine matrix and irreducible polynomial. The conception of confusion is applied to the texture of S-box. In the diffusion phase, the permuted bytes is swapped. The encryption step are done by encrypting an image bytes by using an S-box for substitution and applies nonlinear transformations depending on the values parsed from input fields. And then swapping it by using swap function to permute it. Every pixel is transformed independently, and the new ciphered image is created as a result of these operations. The analysis of correlation coefficient shows that the suggested algorithm effectively removes the correlation between the adjacent pixels. The ciphered image exhibits correlation coefficients near to zero in all the directions, demonstrating strong security and high randomness against the statistical attacks. The results of simulation and security analysis present that the proposed image encryption scheme has high encryption efficiency and successfully encrypts and decrypts diverse types of images. It has kindly key space and can endure statistical, brutal and diverse attacks.

Abstract— The explosion of mobile data traffic and emerging application scenarios (e.g., IoT, immersive multimedia/application, tactile internet) has propelled a drastic evolution from traditional 4G(long-term evolution(LTE)) networks. 5G) mobile system, is believed as no longer an evolution of 4G but a revolutionary industry driven and massively expanded service ecosystem that will enable us to connect the entire crowded world for more than 7 billion peopl world without any limitations and can constitute a "World Wide Wireless Web" (WWWW). The purpose of this paper is to give a well systematic overview of the 5G contributing in application areas for future communication networks. We analyze this transition, from 1G to 4G, with respect to the dynamics of RA technical systems settings, data rate and network technology. The essence of this paper is in depicting the enablers of 5G which include new network structures( hence reconfigurable multi technology cores and cognitive terminals)and game -changer physical layer technologies(hence Full- Duplex communication, Filter Bank Multi- Carrier crisis(FBMC), Non Orthogonal Multiple Access Noma).Sparce Code Multiple Access(Scma). We also comment on major enablers and supporting key technologies including SDR, CR,nanotech-nology, cloud computing to realize the 5G vision.. The study concludes that 5G is the preeminent technology for a digital society and resulted from naturally evolving technical specifications such as higher capacity, lower latency, faster..
Keywords—5G, Network Architecture, Physical Layer, NOMA, SCMA, Full Duplex, FBMC, Cognitive Radio, Software Defined Radio, IoT.

Denoising is a challenging problem within the area of low-level computer vision due to the trade-off between preserving image details and removing noise. While deep learning methods have achieved remarkable results in image denoising, the majority of current approaches work either in the spatial or frequency domain of an image. In this paper, we propose a novel Self-Adaptive Frequency-Spatial Fusion framework that fuses the spatial domain convolutional features with the frequency domain transform coefficients using an edge-aware weight estimation strategy. Our framework consists of two main branches: one that extracts spatial domain convolutional neural network features and another that calculates the frequency domain features using the discrete cosine transform. An edge-aware weight estimation module determines the importance of each of these feature domains at each pixel in the image during the denoising process while preserving image details. Furthermore, a self-adaptive strategy optimally fuses these features together. The results of our framework exhibit significant improvements over current state-of-the-art image denoising approaches on multiple benchmark datasets with an average PSNR improvement of between 0.8 to 1.2 dB. The model also has low computational complexity with only 18.5M parameters and 35.2G FLOPs. Furthermore, the framework generalizes well to different noise levels and types. The results of our framework exhibit a new paradigm in image denosing by fusing the features extracted from the spatial and frequency domain of an image.

Abstract
The Internet of Things (IoT) paradigm enables data transmission for multiple interconnected smart entities and requirements for fundamental and mission benefits across various sectors. IoT networks usually operate Routing Protocols for Low-Power and Lossy Networks (RPL). Nevertheless, RPL lacks protection com-ponents via configuration, presenting IoT-RPL as predisposed to low-overhead internal multiple types of attacks. Normal and attack traffic are usually similar, causing detection challenges for anomalies depending on signature-based intrusion detection systems (IDS). Additionally, a traditional explanation of the correctness of IDS techniques needs to be improved. This proposed a hybrid model of a one-dimensional convolutional neural network and decision tree (HODCNN-DT) approach. The experimental results show the proposed HODCNN-DT attack detection method's effectiveness. This approach enhances detection's precision, recall, and accuracy compared to other techniques.

Abstract: Over the last decade, image steganalysis, the science of detecting hidden information in digital images, has undergone a fundamental transformation, from manually engineered statistical feature methods to sophisticated deep learning architectures that can automatically extract features. This survey presents a systematic and detailed review of image steganalysis techniques, which can be grouped into four main categories: 1) Traditional intensity and texture-based methods including threshold-based methods and Grey Level Co-occurrence Matrix (GLCM) features; 2) Handcrafted feature-based methods such as the Spatial Rich Model (SRM), SPAM, DCTR, and GFR with ensemble classifiers; 3) Machine learning and deep learning methods including CNN-based architectures such as Xu-Net, Ye-Net, Yedroudj-Net, and SRNet; and 4) Hybrid fusion-based frameworks combining handcrafted and learned features to enhance detection robustness. The survey also includes common evaluation criteria, popular benchmark datasets such as BOSSbase, ALASKA II and IStego100K, and a comparison of stated detection accuracies in terms of methodologies and datasets. Key findings indicate that deep learning approaches consistently outperform traditional methods, with reported accuracies of up to 99% on some benchmarks, but important challenges remain unaddressed, such as the cover-source mismatch, low-payload detection, and the generalization gap between controlled laboratory conditions and real-world social media transmission. This review highlights these outstanding difficulties and discusses the future research directions that need to be followed in order to produce universal, resilient and computationally efficient steganalysis systems that can work efficiently in realistic deployment situations.

Abstract: Image inpainting, one of the main tasks of image restoration that fills missing or hidden portions across an image while maintaining structural correlations and perceptual realism. As real-world occlusions are more complex, conventional image restoration methods have been not sufficient anymore and thus let themselves be widely replaced by modern deep learning–based approaches. Recent developments have shown that large missing areas, structured textures and semantic coherence are better modeled by data-driven methods improving performance in challenging sub-problems like facial image restoration. This paper presents a comprehensive survey of the current state-of-the-art in image inpainting approaches and special attention is drawn on deep learning–based methods. Here, detailed deep learning–based methods are systematically classified into three main categories: GANs, transformer-based models and diffusion models.The elemental principles, architectural features and restoration abilities for every class are reviewed along with the progress of image inpainting methods from recent years.This review also provides summaries of their prominent quantitative evaluation metrics from a pixel-based and perceptual perspective, along with benchmark datasets commonly used in literature as well as the most frequently adopted mask types for image inpainting.We provide a comparative study of representative modern methods, which is intended to enable objective evaluation over various datasets and metrics. Finally, the paper summarizes challenges and future research opportunities on open topics like structure-texture trade-off, perceptual quality assessment and robustness against heavy occlusion. The purpose of this review is to equip researchers with a well-organized and current perspective on contemporary advancements in image inpainting research while providing useful references for future endeavors.