Fusion: Practice and Applications

There is a need to create and develop smart cities that could help improve the quality of life in global countries. The goal of this paper is to develop a novel smart city framework for the GCC countries. This study presents a comprehensive analysis of smart city features across multiple cities worldwide, leveraging data from a reliable world cities database. Through exploratory data analysis and visualization techniques, we examined various aspects of smart city development, including mobility, environment, government, economy, people, and living standards. It turned out from the literature that globally, there is a focus on some of the dimensions of smart cities while others did not receive much attention. Smart economy and smart environment were not receiving much attention globally. A framework was developed for the GCC countries that focuses on all the dimensions of the smart cities, but most of the attention is on smart governance and smart economy since these two dimensions help improve the quality of life and diversify the sources of the economy in the country. This framework is useful for GCC countries as it would have great implications on the desired outcomes of smart cities and link with the strategic development goals that most GCC countries have, whether it is the 2030, 2035, or even the 2040 vision.

Pancreatic cancer (PC) is an extremely malignant cancer type with a maximum rate of mortality. It remains a challenging form of tumor to treat due to its late analysis and aggressive nature, which drastically decreases the survival rate. Early analysis of PC is vital for enhancing the probabilities of treatment and survival. PC analysis was initially dependent upon imaging, and then the recent imaging offered a worse prognosis, restraining clinicians’ treatment choices. PC detection utilizing deep learning (DL) contains the application of advanced computational methods for analyzing medical image data like CT scans or MRI images, for the early and correct detection of PCs. DL approaches, particularly convolutional neural networks (CNNs), are trained on huge databases for diagnosing forms and anomalies indicative of PC. Therefore, this study presents a novel Coot Optimization Algorithm with Deep Learning based Medical Image Analysis for Pancreatic Cancer Diagnosis (COADL-MIAPCD) technique. The main objective of the COADL-MIAPCD approach is to proficiently examine the medical images for the detection of PC. The COADL-MIAPCD technique primarily applies a median filtering (MF) for image pre-processing. In addition, the COADL-MIAPCD approach allowed using of an improved SE-ResNet. Moreover, the COA has been utilized for the optimum parameter choice of the improved SE-ResNet. At last, the extreme learning machine (ELM) has been used for the recognition and classification of PCs. The simulation outcomes of the COADL-MIAPCD technique has been validated utilizing a medical image database. The obtained experimental values stated that COADL-MIAPCD technique achieves better performance than other models.

Physical characteristics of underwater environments, such as absorption, scattering, and progressive color loss, are only a few of the many factors that cause underwater images to degrade. Additionally, the turbid water and marine plankton influence the degradation of these images. All of these play a major role in the difficulty of extracting features from underwater images. This study aims to develop a new system that combines color correction techniques and weight maps to address the challenges caused by underwater environments. First step: the color correction, which consists of both color compensation and white balance, are used to improve the colors of images. In the second step:  a comprehensive enhancement solution has been adopted on the two images that resulted from the first step by performing two different ways, the first image is improved by an image sharpening algorithm, and the gamma correction is used to process the second one. Four weights maps are applied for feature extraction and finally multi-scale fusion process is used to find the final enhanced image. Three types of underwater scenes are used (Bluish, Greenish and Foggy) to assess the suggested work. In addition to evaluating the results visually, a number of statistical metrics (IE, PCQI, AG, UIQM and UCIQE) are used to evaluate the results and compare them with previous works. The results indicate a marked improvement in all types of image.

Weather monitoring is a vital challenge in dissimilar areas of applications such as military missions, higher precision agriculture, outdoor entertainment and recreation, industrial manufacture, and logistics. The most vital application is natural weather disaster monitoring. Weather change has made stronger an occurrence of natural disasters all over the world. More extreme climate events have been experienced for the past few years, like lower and higher temperatures, sturdy winds in humid cyclones, heavy rains, and intensified lack. Therefore, at present, remote sensing imagery (RSI) analysis is necessary in the field of ecological and weather monitoring mainly for the application of identifying and handling a natural climate disaster. To upsurge the accuracy of detection, machine learning (ML) and deep learning (DL) systems were applied to enhance the efficacy of removing features and help to perceive large-scale losses like landslides, earthquakes, and floods. In this manuscript, we design and develop a Weather Disaster Detection Model Using Zebra Optimization Algorithm with Ensemble Learning on Remote Sensing Images (WDDZOA-ELRSI) technique. The proposed WDDZOA-ELRSI model's main intention is to improve the detection model of weather disasters using state-of-the-art DL methods. Initially, the bilateral filter (BF) method is employed in the image pre-processing stage to eliminate the unwanted noise from input data. Furthermore, the feature extraction method executes GoogleNet technique to transform raw data into a reduced set of relevant features. For the classification process, the ensemble of deep learning models such as conditional variational autoencoder (CVAE), graph convolutional network (GCN), and Elman recurrent neural network (ERNN) have been deployed. Eventually, the zebra optimization algorithm (ZOA)-based hyperparameter tuning procedure has been achieved to improve the detection outcomes of ensemble models. The simulation analysis of the WDDZOA-ELRSI system is verified on a benchmark image dataset and the outcomes were evaluated under numerous measures. The simulation outcome emphasized the enhancement of the WDDZOA-ELRSI model in the weather disaster detection process

Wireless Sensor Network (WSN) signifies a state-of-the-art technology that combines energy-effective sensors with wireless transmission services enabling prompt surveillance and data collecting from the nearby environments. Owing to the intrinsic features of WSNs, they face numerous challenges of security that range from resource-based attacks, like computational overload or energy depletion, to interception, eavesdropping, and tampering. With the hacked data, the attackers can replicate the same sensors and use clones in the corresponding WSNs. This kind of cloning of the sensors, which is comprised of the WSN, is called a clone attack. Since the replicated sensors formed by the attackers have parallel keys and information, therefore the clone attacks have become a great attack for WSN. To defend WSNs against cyberattacks, machine learning (ML) and deep learning (DL) were applied to classify malicious and normal traffic. This study designs an Attack Detection and Mitigation using Deep Learning with an Optimization Algorithm in Wireless Sensor Networks (ADMDL-OAWSN). The main objective of the ADMDL-OAWSN system is to improve security in cloned nodes for the cyberattack detection model. In the primary step, the data pre-processing employs the StandardScalar method to transform input data into a suitable format. Next, the proposed ADMDL-OAWSN model designs a crayfish optimization algorithm (COA) for the subset of the feature selection (FS) to pick the most related features from an input dataset. For the attack classification process, the convolutional neural network and bi-directional gated recurrent unit with attention mechanism (CNN-BiGRU-A) technique have been exploited. At last, the parameter tuning of the CNN-BiGRU-A is applied by the design of the secretary wolf bird optimization (SeWBO) algorithm. Extensive experiments have been conducted to validate the results of the ADMDL-OAWSN system. The simulation results revealed that the ADMDL-OAWSN system emphasized furtherance when compared to other recent systems

Cyber-physical systems (CPSs) unite the computation with physical methods. Embedded networks and computers observe and handle the physical procedures, generally with feedback encircles whereas physical procedures affect computation and conversely. In the last decade, the prompt growth of network-associated services has formed confidential information on the Internet. However, networks are much inclined to intrusions wherever unapproved consumers try to retrieve confidential data and even disturb the systems. Constructing a proficient network intrusion detection system (IDS) can be essential to avert these attacks. Utilizing digital twin technology enhances the IDS of physical devices in CPSs. IDSs normally utilize machine learning (ML) techniques for categorizing the attacks. However, the features employed for classifications are not appropriate or adequate all the time. Moreover, the amount of intrusions can be significantly lower than the amount of non-intrusions. Therefore, simple techniques may fail to deliver satisfactory performances owing to this class imbalance. In this study, we offer a Metaheuristic-Driven Hybrid Deep Learning Model for Robust Intrusion Detection in Secure Cyber-Physical Systems (MHDLM-RIDCPS) model in Smart City Environment. The proposed MHDLM-RIDCPS technique primarily targets the classification and recognition of intrusions using digital twin technology to enhance security within the CPS. Primarily, the proposed MHDLM-RIDCPS approach utilizes min-max normalization for transforming an input data into a standardized format. To alleviate dimensionality issues, the coyote optimization algorithm (COA) can be executed to select a subset of features. In addition, the modified prairie dog optimizer (mPDO) combined with a convolutional neural network and bi-directional long short-term memory with attention mechanism (AM-CNN+BiLSTM) classifier is exploited for the identification of intrusions. The design of the mPDO system primarily concentrates on the parameter optimizer of the AM-CNN+BiLSTM algorithm and so improves the classifier performances. To determine the greater efficiency of the MHDLM-RIDCPS system, a comprehensive set of simulations can be applied and the performances are tested over distinct aspects. The experimental analysis guaranteed the superior results of the MHDLM-RIDCPS methodology with existing methods

The objective of the presented work is the examination of ergonomic challenges of prolonged sitting in digital learning using an instrumental multi-criteria decision-making technique named 'TOPSIS' (Technique for Order of Preference by Similarity to Ideal Solution). A total of sixteen ergonomic challenges of prolonged sitting in digital learning have been identified by a group dialogue with laptop, tablet, smartphone users, academicians, and students. The study compares equal weight ages and variable weight ages, finding that eye strain, neck pain, and mental tiredness are the most close to ideal solutions, while leg pain is the least. Linear Reggression, a machine learning approach, is the best-performing model, with Neural Network and SVM showing marginal improvement. The outcomes of the experiment demonstrate that the suggested model functions well in terms of accuracy, and techniques have been used to raise the diagnostic rate and solve the issue. The outcomes can be very helpful in finding and applying measures to deal with ergonomic challenges of prolonged sitting in digital learning. Policymakers may use the output of this study regarding the relative importance and productivity influencing tendency of these chosen sixteen ergonomic challenges, for creating mechanisms for the betterment of human-computer interface. 

Accurate emotion detection is crucial for individuals facing communication barriers, yet existing approaches struggle with real-time limitations and information Individual privacy. This research presents a new IoT-based framework that integrates EEG and physiological signals from wearable sensors with deep learning models, including CNN, Decision Trees, SVM, KNN, and Naïve Bayes. Unlike traditional methods, our approach effectively mitigates data latency and sensor noise while ensuring compliance with GDPR and HIPAA standards. Experimental results demonstrate a validated accuracy of 99-100%, outperforming state-of-the-art models. These developments establish our framework as a game-changing instrument for affective computing applications, enhancing human-machine interaction and healthcare quality of life.

Blockchain technology is now widely used in data sharing, cryptocurrency industry, Internet of Things and other fields. However, despite its increasing use, security and privacy concerns remain important issues. Blockchain security is enhanced by the use of hashing algorithms that ensure data integrity and provide a solution to security problems, but hashing algorithms usually have limitations in terms of resource consumption, memory and speed. To overcome these obstacles, the efficiency and security of the hashing algorithm used in blockchain must be increased. This paper presents a proposal to improve the hashing process in blockchain by leveraging the lightweight quantum algorithm Ascon, which has been improved after integrating it with nonlinear Lotka-Volterra equations. This integration can improve performance and security by combining the mathematical principles of these nonlinear equations to study the interactions between systems. Through this integration, it is possible to improve power management and work on intelligent resource allocation, as well as make the system more robust against attacks by complicating the random number generation process. The performance of the proposed system was tested in terms of throughput, elapsed time, amount of memory used, and time required to process data. The results showed that the proposed algorithm outperforms the original Ascon algorithm in terms of providing faster processing while maintaining a high level of performance and security, reducing time, and increasing the amount of data processed with less memory required for storage. These improvements are of great importance in developing blockchain technology and enabling its multiple uses in many applications. 

Medical imaging performs a critical position in modern healthcare, in particular in the early detection of cancers, which considerably enhances survival charges and treatment consequences. This study investigates a hybrid version combining Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) to optimize medical image analysis. Leveraging advanced deep gaining knowledge of strategies along with Transfer Learning and Data Augmentation, the hybrid method validated advanced performance in class, segmentation, and anomaly detection obligations. Experimental results discovered that the hybrid version outperformed standalone CNN and ViT architectures, attaining high diagnostic accuracy whilst keeping computational efficiency. The findings spotlight the potential of AI-stronger answers to revolutionize clinical diagnostics by way of offering accurate and reliable computerized systems, paving the manner for broader medical programs and improved patient results.

Artificial Intelligence (AI) can become a great asset to produce cryptographic keys in order to improve the security of the encryption methods. While using machine learning algorithms AI can generate most complex and unpredictable keys to prevent brute-force and cryptanalyst attacks. Key generation using AI also allows the design of cryptographic solutions that adapt to the context in which the key is used. It also enhances the conventional security measures while simultaneously providing great opportunities for creating flexible security solutions. This paper proposed a new text security method based on the integration of the cryptography and steganography, where the suggested method is done based on OpenAI API. The proposed method is consisted of three steps, and these steps are key generation, text encryption, and data embedding. The first step, is utilized by using GPT-2 model to generate set of keys for both cryptography and steganography steps. The second step, is starting by converting the plaintext to ASCII format, then performed modulo arithmetic operation between ASCII values and the keys that generated from the previous step, then convert the achieved equation results to Hexadecimal format, and finally convert these values to binary and these values represent the final ciphertext. The last step of the proposed method is done by hiding the binary values within image, this done by select positions randomly, then used GPT-2 model to generate another set of keys to shift the values of random positions, then applied least significant bit (LSB) algorithm to hide the bits within the final position with different color channels. The proposed approach provides a basis for the development of new-generation secure communication systems in the context of AI.

Lately, handling big data has become challenging due to its large size and complexity. To address scalability, availability, real-time performance, flexibility, and various Quality of Service (QoS) characteristics, Data Distribution Services (DDS) middleware offers extensive integration with mission-critical, real-time, and high-performance networks. Unlike traditional client-server communication models, Data Distribution Services rely on a publish/subscribe communication model. DDS enhances the quality of video streaming through its efficient data delivery approach. On internet protocols, a significant portion of traffic is generated by content delivery applications, such as video streaming. This study examines how Data Distribution Services are well suited for streaming real-time, full-motion videos over communication networks. Several experimental studies have been conducted to compare video streaming using a VLC player with an overlay of Data Distribution Services. Our application-aware routing system enables mobile network operators to utilize their networks more efficiently, allows service providers to improve customer satisfaction, and ensures end-users experience desirable service quality across various network applications. The findings of this study demonstrate the efficiency of DDS in delivering high-quality video streams while utilizing low network bandwidth. Additionally, the results highlight that DDS offers greater flexibility and scalability, making it a highly important technology for video distribution over internet protocol networks. It achieves this by using narrower bandwidth while maintaining high-quality video stream delivery.

Wireless Sensor Networks are everywhere around us used in variety of applications such as weather forecasting, military surveillance, health monitoring, agriculture monitoring, and smart IoTs etc. These networks are particularly employed to sense and broadcast the data from source nodes to sink node. Hence, energy consumption becomes one of the most challenging jobs here. Hierarchical clustering-based routing schemes prove to be helpful in such situations. As a result, optimized cluster head selection is essential and key task here. In this paper author has attempted to design an optimized cluster head selection scheme based on Adaptive Hybrid Dragonfly Firefly (AHDF) algorithm based on node energy, corresponding distance and network load and delay parameters. The simulation and comparison results showcase the outperformance of the proposed routing scheme in terms of energy efficiency (121% and 41%), network lifetime (89% and 21%) and data throughput (31% and 23%) in comparison of existing routing schemes SEELCA [15] and CRCGA [16] respectively.

Ransomware or crypto-ransomware is a big headache to digital media and transactions nowadays. Generally, Ransomware affects the operating system and transfers the valuable information and data stored in the system. Some ransomware attacks the system and corrupts the system file, making it useless to the user. Data encryption with a private key is also one of the attaching fashions of some types of ransomwares. Most ransomware attacks are reported in android operating system-based devices. The solution to ransomware is only the earlier identification of an attacked pattern in the operating system and removal of it. Artificial Intelligence (AI) plays a major role in various kinds of attack detection and classification processes. Machine learning (ML) technique can be used to train and classify the presence of ransomware in android-based devices. Various parameters, such as the characteristics of applications' permission access to various inputs of the devices. The data can be used to train the Recurrent Neural Network (RNN), the most popular and highly accurate ML module that performs a highly accurate classification process. The performance can be evaluated using various sensitivity evaluation metrics such as accuracy, sensitivity, specificity, and precision.

This study proposes the novel machine learning concepts to enhance both prediction accuracy of diabetes detection and interpretation of diagnostic models. First, the methodology uses multiple imputations by chained equations (MICE) to complete data before analysis through missing data imputation procedures. The class imbalance problem is solved through the implementation of Synthetic Minority Over-sampling Technique (SMOTE). The Interquartile Range (IQR) outlier detection method helps remove outliers because it enhances model robustness. The hybrid RFE-WWO selection process combines Recursive Feature Elimination (RFE) with Water Wave optimization (WWO) to select important features that strike the right balance between model complexity and prediction accuracy. The HFM framework contains the Hybrid Fusion Model as its essential component, which merges AdaBoost's and CatBoost's most favorable aspects. The hyperparameter optimization with TPE leads to model tuning which reaches a prediction accuracy of 97.84% through the application of Tree-Structured Parzen Estimator. The entire approach delivers enhanced accuracy and it improves precision along with recall metrics and F1 score performance of the predictive model. The framework shows significant potential for early diagnosis by merging these advanced techniques since ensemble methods are essential for healthcare data analysis while accurate interpretable models are vital to create dependable diagnostic tools.