Abstract Smart home networks are a rapidly evolving part of the Internet of Things (IoT), where connected devices exchange information to automate energy use, enhance comfort, and improve user safety. Despite their benefits, these systems face ongoing challenges related to data privacy, communication security, and eco-friendly energy management, especially when centralized control introduces risks such as single points of failure and delays. Existing solutions, including cloud-based encryption, blockchain-powered energy trading, and lightweight authentication, have improved trust and interoperability but still encounter issues related to scalability, high processing demands, and limited resistance to quantum threats. To address these limitations, this paper presents a blockchain-based system that enhances privacy, security, and energy efficiency in smart homes. The system integrates Machine-to-Machine (M2M) communication and IoT protocols are utilised and optimised within a decentralized architecture composed of communication, security, and application layers, making it modular and reliable. A flexible, post-quantum key exchange method based on the Supersingular Isogeny Diffie-Hellman (SIDH) protocol ensures device and user authentication. The system utilises a range of communication protocols, including WebSocket, UDP, and TCP, within the decentralised architecture to balance reliability and performance across different smart home applications. The design also introduces distinct access levels for users and administrators to strengthen control and support collaboration among neighbouring homes for collective energy savings. Tests conducted using Cisco Packet Tracer demonstrate notable performance improvements: approximately a 22% reduction in latency, an 18% enhancement in energy efficiency, and a 15% increase in throughput compared to traditional blockchain-IoT configurations. These results indicate that the proposed system provides a secure, scalable, and energy-efficient framework for the future of smart home networks.

The rapid expansion of biometric authentication technologies worldwide has heightened the need for highly reliable and secure identification methods. This research explores the human ear as a distinctive biometric trait, capitalizing on its stable and person-specific anatomical structure. Although ear biometrics offer notable advantages, their practical use is hindered by image variations arising from changes in pose, scale, rotation, illumination, and contrast. To overcome these challenges, this paper presents an innovative deep learning-based ear recognition framework. The proposed approach employs a dual-stream feature extraction strategy that integrates two advanced Convolutional Neural Network (CNN) models MobileNetV3 and DenseNet-121 to derive rich and complementary feature representations, which are subsequently fused. The resulting high-dimensional feature space is then optimized using a Multi-Learning Strategy Golden Eagle Optimization (MLSGEO) algorithm to retain only the most discriminative features. To strengthen security and privacy, the refined feature vector is transformed into a non-invertible, cancelable biometric template using a Comb-filter-based protection mechanism. Data augmentation techniques are further applied to compensate for dataset size limitations. The framework was evaluated on five benchmark ear datasets: AMI, AWE, IITD-I, IITD-II, and UERC, achieving recognition accuracies of 99.90%, 99.64%, 99.78%, 99.32%, and 93.31%, respectively. Experimental findings show that the proposed system outperforms existing state-of-the-art methods. Overall, the integration of robust feature learning with a resilient template protection scheme demonstrates strong potential for secure and high-accuracy biometric authentication applications.

Optical signal processing plays a crucial role in Intrusion Detection System (IDS) for optical networks, as optical techniques provide more bandwidth and more security for the communication channel. It involves various techniques and operations performed on the optical signals to extract relevant information and identify potential security threats or intrusions. The IDS is a very important component which detect unauthorized access to computer networks and systems. IDS analyzes network traffic and system logs to identify potential security threats and alert system administrators to take appropriate action. There are different types of IDS, including signature-based, anomaly-based, and hybrid approaches. IDS uses a variety of detection methods to identify potential security threats, including statistical analysis, machine learning, and rule-based methods. However, IDS faces several challenges, including the need to balance detection accuracy with false positives and false negatives while keeping up with the ever-changing threat landscape. There are an evaluation metrics, such as detection rate, false positive rate, and accuracy. In this paper, we use accuracy of the detection to evaluate the effectiveness of IDS. Data preprocessing and feature selection are also important to improve search capabilities in IDS. In general. Ongoing research in this area, including the use of deep learning models such as convolution neural network (CNN), long-short term memory (LSTM) and hybrid models with CIC-IDS 2018 dataset, will continue to improve IDS ability to detect and prevent security breaches. This research will enhance accuracy detection of the IDS.

The rapid proliferation of Internet of Things (IoT) devices has expanded the attack surface of modern networks, necessitating the development of more robust and intelligent Intrusion Detection Systems (IDS). In this study, we propose two hybrid deep learning models tailored for multiclass intrusion detection on BoT-IoT datasets. The first model combines XGBoost for efficient feature extraction with a Recurrent Neural Network (RNN) to capture temporal dependencies in attack sequences. The second model leverages a Transformer architecture for deep contextual learning of major attack classes and LightGBM for handling class imbalance among minor classes. Extensive experiments conducted on the BoTNeTIoT-L01-v2 and NF-Bot-IoT datasets demonstrate that both proposed models significantly outperform existing state-of-the-art methods. The XGBoost-RNN model achieves high classification performance over all metrics with an accuracy of 99.99%, while the Transformer-LightGBM model consistently records 99.45% across all key evaluation metrics with the highest F1-score. These results highlight the effectiveness of the hybrid architectures in enhancing detection accuracy, class balance, and adaptability in complex IoT threat landscapes.

Blockchain technology offers a robust framework for integration with the Internet of Things (IoT), enhancing interoperability, security, privacy, and scalability in modern technological ecosystems. However, traditional cryptographic protocols used in blockchain systems are increasingly vulnerable to quantum attacks due to advancements in quantum computing. In response, the National Institute of Standards and Technology (NIST) has prioritized research in post-quantum cryptography, presenting challenges and opportunities for developing blockchain-based applications tailored to IoT devices. Among the post-quantum cryptographic schemes evaluated in NIST's third standardization round, the Supersingular Isogeny Key Encapsulation (SIKE) protocol stands out for its relatively small public and private key sizes. Despite this advantage, SIKE faces challenges related to high latency, necessitating efficient implementations to make it viable for real-world applications. This research focuses on optimizing the cryptographic foundations of blockchain networks to securely and efficiently integrate resource-constrained IoT ecosystems. By enhancing the SIKE protocol, which exhibits strong resistance to brute-force and whitewashing attacks, the study achieves significant performance improvements. Our FPGA-based implementation on the VIRTEX-6 XC6VLX760 demonstrates reduced latency, achieving a key generation time of 24 ms, encapsulation time of 72 ms, and decapsulation time of 73 ms for SIKEp434. These results highlight the feasibility of deploying SIKE-optimized blockchain networks in IoT environments with stringent resource constraints.

In the evolving landscape of Speaker Recognition (SR), the majority of research traditionally focuses on scenarios devoid of ambient noise or interference, largely adhering to text-dependent protocols, where the identity is confirmed through specific verbal cues. This approach, while methodologically efficient under controlled conditions, starkly contrasts with the complex and unpredictable nature of real-world environments in the presence of various acoustic disturbances. Addressing this gap, our study is concerned with the exploration of text-independent SR under realistic interference conditions, aiming to authenticate speakers based on the intrinsic characteristics of their speech, independent of the linguistic content conveyed. Central to our methodology is the strategic employment of advanced signal separation, serving as a critical pre-processing step to reduce the impact of external noise and interference. This work facilitates a more accurate and efficient identification process, pivotal for text-independent SR systems. Our framework harnesses the computational prowess of two leading-edge Deep Learning (DL) architectures: the Long-Short Term Memory Recurrent Neural Network (LSTM-RNN) and the Deep Convolutional Neural Network (CNN). Through a meticulous comparative analysis, we evaluate these models' efficacy in various interference landscapes, benchmarking their performance against established conventional systems. A novel aspect of our research is the application of signal separation in different domains, yielding significant enhancements in SR accuracy, notably within the time domain. Empirical results of experiments reveal that a three-layer CNN configuration achieves a high recognition rate of 97.33

This paper presents a blockchain-based framework designed to enhance privacy, security, and sustainable energy management in smart home environments. The system integrates Machine-to-Machine (M2M) communication and Internet of Things (IoT) protocols within a decentralized framework to facilitate secure, real-time interactions between devices and users. It features a multi-layered design comprising communication, security, and application layers, ensuring dependable functionality and modularity. The Supersingular Isogeny Diffie-Hellman (SIDH) protocol provides security for device-to-device and user interactions with post-quantum cryptographic protection. System behavior is simulated using Cisco Packet Tracer, and additional protocols, such as WebSocket (via websocat), UDP (User Datagram Protocol), and TCP (Transmission Control Protocol), are tested to enhance interactivity and communication efficiency. The architecture assigns distinct roles to users and administrators, establishing clear access and control privileges. Additionally, the framework explores the coordinated optimization of sustainable energy use among neighbouring smart homes. Key performance indicators include system security, latency, communication throughput, and energy efficiency. This proposed model seeks to offer a secure and scalable solution for the future of smart home systems.

This paper is mainly concerned with video watermarking as a tool to secure the video transmission process over wireless channels. In addition, the relatability of the communication process is guaranteed through a hybrid error control scheme. The watermarking depends on applying a hybrid structure of Block-based Singular Value Decomposition (B-SVD) and SVD schemes. The rationale behind the utilization of SVD for video watermarking is the fact that singular values of a frame are not severely affected by noise or disturbance induced on that frame. The error control is performed with an efficient hybrid post-processing scheme, which comprises Spatial Circular-Scan Order Interpolation Algorithm (CSOIA), temporal Partitioning Motion Compensation Algorithm (PMCA) and Bayesian Kalman Filter (BKF). In the proposed framework, cornea and infrared frames are watermarked using two stages of SVD watermarking. The watermarking scheme includes embedding and extraction stages. Two watermark images are embedded in the cornea and infrared frames in the embedding stage using the hybrid structure of B-SVD and SVD schemes. Next, the watermarked cornea or infrared frames are transmitted through the erroneous wireless channel. The received corrupted cornea or infrared frames are recovered using the proposed post-processing error control schemes. Finally, the inverse process of hybrid SVD and B-SVD is employed in the watermark extraction stage. Simulation results for several cornea and infrared frames show that the proposed framework has extremely adequate subjective and objective video quality metrics compared to the traditional methods. In addition, the watermark robustness, security, and detectability are enhanced. Moreover, the experimental results clarify that the proposed watermarking scheme is superior and more secure than the other previous schemes for embedding and extracting watermarks efficiently in the presence of attacks.

This paper presents two efficient approaches for object detection from Infrared (IR) images. The first approach is based on enhancement using histogram equalization (HE) in addition to gradient estimation using Laplacian filter, and finally estimation of the cumulative histogram for the detection or classification task. The second approach is based on gradient estimation after a hybrid structure comprising HE and contrast-limited adaptive histogram equalization (CLAHE), and finally cumulative histogram estimation for object discrimination. After the cumulative histogram estimation, the difference between cumulative histograms with and without objects is estimated and used for discrimination in object detection or gait recognition. The two proposed approaches in this paper are compared with traditional ones including CLAHE with gradient and cumulative histogram estimation (CLAHE-GCH); additive wavelet transform with homomorphic processing, gradient and cumulative histogram estimation (AWTH-GCH); and histogram matching with gradient and cumulative histogram estimation (HM-GCH). Simulation results prove the necessity of using an enhancement technique as the first step in detecting objects from IR images. The results demonstrate also the success of both proposed approaches in detecting objects from IR images compared to traditional methods.

It is imperative to note that post-quantum cryptography, such as supersingular isogeny Diffie-Hellman (SIDH), is essential for ensuring that Internet of Things (IoT) devices have a restricted amount of resources. The primary challenges in assuring the security of IoT devices are addressed by this work's analysis of SIDH implementations designed for field programmable gate array (FPGA) architecture. Extensive efforts towards implementing the architecture for a rapid and constant-time FPGA implementation of SIDH. This quantum-resistant cryptographic primitive is a crucial component for adhering to NIST's PQC standardization. Our goal with this paper is to demonstrate how FPGA architectures can enhance the parallelism of SIDH, making it a more practical option for securing resource-limited IoT devices. Our focus is on providing a reliable speed record for SIDH, which is crucial for ensuring the security of IoT devices. The design for isogeny computation was built over p434 in Xilinx Virtex 6 and produced 45ms for public key generation in addition to 35ms for secret key generation.

Nowadays, biometric systems have replaced password-or token-based authentication systems in many fields to improve the security level. However, biometric systems are also vulnerable to security threats. Unlike passwords, biometric templates cannot be replaced if lost or compromised. To deal with issue of compromising biometric templates, template protection schemes have evolved to make it possible to replace the biometric templates. A cancellable biometric scheme is such a template protection scheme that can replace a biometric template, when it is stolen or lost. The biometric used here is speech. It is important to preserve user confidentiality. Cancellable biometrics is a new notion addressed for this problem. This paper presents a scheme for cancellable speaker recognition based on spectrogram patch selection. The simulation results reveal that the suggested approach is practical, and it satisfies the desired criteria such as renewability, security and performance. The accuracy of the proposed cancellable speaker recognition scheme reaches 98.75% with a Convolutional Neural Network (CNN) composed of three layers.

The main characteristic of deep learning approaches is the ability to learn differentiating and discriminating features.These techniques can discover complex relations and structures within high-dimensional data.For feature extraction, deep learning models employ several layers of nonlinear processing units.One of the fields that have applied deep architectures with a noticeable breakthrough in performance measures is Natural Language Processing (NLP).Recurrent neural networks (RNNs) and their variants Long-Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) are commonly used for NLP applications as they are efficient at processing sequential data.Unlike RNNs, LSTMs and GRUs can combat vanishing and exploding gradients.In Addition, Convolutional Neural Network (CNN) is another deep architecture that has been widely used in language processing.On the other side, sentiment analysis (SA) is an NLP task concerned with opinions, attitudes, emotions, and feelings.Sentiment analysis deduces the author's attitude regarding a topic and classifies the attitude polarity according to a set of predefined classes.Application of SA in business analytics helps to gain insight into consumer behaviour and needs.In the proposed work deep LSTM, GRU, and CNN are applied for Arabic sentiment analysis.The models are implemented and tested employing character-level representation.Also, deep hybrid models that combine multiple layers of CNN with LSTM or GRU are studied.The application aims at investigating the capability of deep LSTM, GRU, and hybrid architectures to learn and extract features from characterlevel representation.Results show that combining different architectures can boost performance in SA tasks.The CNN-LSTM/GRU combinations registered higher accuracy compared to deep LSTM and GRU.

Biometric recognition is an automated technique of recognising persons based on their traits. Because of their exceptional texture, the biometric features' ostensibly random nature makes them good candidates for recognition. These features are unique for each individual even for identical twins authentication. The latest developments in Deep Learning (DL) and computer vision has proved that Convolutional Neural Networks (CNNs) can extract generic descriptors that can represent complex image features. How to protect the biometric data and ensure user’s privacy is a main concern, nowadays. Hence, several cancelable biometric scenarios have been proposed. In this paper, we propose a novel cancelable biometric recognition system based on a CNN model with bio-convolution. The performance metrics are estimated on different face and iris datasets. In contrary to most conventional secure biometric recognition systems, the proposed system achieves superior accuracy results, while keeping the ability to cancel the biometric traits if compromised. The experimental findings on each database are shown and compared to those of the state-of-the-art systems that have been tested on the same database. Furthermore, the recognition rates reach 99.15%, 98.35%, 97.89, and 95.48% with the LFW, FERET, IITD, and CASIA-IrisV3 databases, respectively.

Automatic Speaker Recognition (ASR) in mismatched conditions is a challenging task, since robust feature extraction and classification techniques are required. Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) is an efficient network that can learn to recognize speakers, text-independently, when the recording circumstances are similar. Unfortunately, when the recording circumstances differ, its performance degrades. In this paper, Radon projection of the spectrograms of speech signals is implemented to get the features, since Radon Transform (RT) has less sensitivity to noise and reverberation conditions. The Radon projection is implemented on the spectrograms of speech signals, and then 2-D Discrete Cosine Transform (DCT) is computed. This technique improves the system recognition accuracy, text-independently with less sensitivity to noise and reverberation effects. The ASR system performance with the proposed features is compared to that of the system that depends on Mel Frequency Cepstral Coefficients (MFCCs) and spectrum features. For noisy utterances at 25 dB, the recognition rate with the proposed feature reaches 80%, while it is 27% and 28% with MFCCs and spectrum, respectively. For reverberant speech, the recognition rate reaches 80.67% with the proposed features, while it reaches 54% and 62.67% with the MFCCs and spectrum, respectively.

Sentiment analysis is a Natural Language Processing (NLP) task concerned with opinions, attitudes, emotions, and feelings. It applies NLP techniques for identifying and detecting personal information from opinionated text. Sentiment analysis deduces the author's perspective regarding a topic and classifies the attitude polarity as positive, negative, or neutral. In the meantime, deep architectures applied to NLP reported a noticeable breakthrough in performance compared to traditional approaches. The outstanding performance of deep architectures is related to their capability to disclose, differentiate and discriminate features captured from large datasets. Recurrent neural networks (RNNs) and their variants Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Bi-directional Long-Short Term Memory (Bi-LSTM), and Bi-directional Gated Recurrent Unit (Bi-GRU) architectures are robust at processing sequential data. They are commonly used for NLP applications as they-unlike RNNs-can combat vanishing and exploding gradients. Also, Convolution Neural Networks (CNNs) were efficiently applied for implicitly detecting features in NLP tasks. In the proposed work, different deep learning architectures composed of LSTM, GRU, Bi-LSTM, and Bi-GRU are used and compared for Arabic sentiment analysis performance improvement. The models are implemented and tested based on the character representation of opinion entries. Moreover, deep hybrid models that combine multiple layers of CNN with LSTM, GRU, Bi-LSTM, and Bi-GRU are also tested. Two datasets are used for the models implementation; the first is a hybrid combined dataset, and the second is the Book Review Arabic Dataset (BRAD). The proposed application proves that character representation can capture morphological and semantic features, and hence it can be employed for text representation in different Arabic language understanding and processing tasks.

You Only Look Once version 3 (YOLOv3) is a deep learning model for object detection and classification. It is a single neural network architecture model that uses features from the feeding images and predicts bounding box for all classes of image simultaneously. This paper descript an experimental work for train the deep learning model based on YOLOv3 architecture implemented using Tensor Flow as a deep learning framework. The training process had been done using the data-set PASCAL VOC 2007 and data-set PASCAL VOC 2012 and using The Adaptive Moment Estimation Optimizer (ADM optimizer). The trained model is then tested by using the VOC 2007 test data-set. The final results evaluate the YOLOv3 deep learning model performance for object detection and classification.

Orbital angular momentum-shift keying (OAM-SK), which is the rapid switching of OAM modes, is vital but seriously impeded by the deficiency of OAM demodulation techniques, particularly when videos are transmitted over the system. Thus, in this paper, 3D chaotic interleaved multi-coded video frames (VFs) are conveyed via an N-OAM-SK free-space optical (FSO) communication system to enhance the reliability and efficiency of video communication. To tackle the defects of the OAM-SK-FSO mechanism, two efficient deep learning (DL) techniques, namely convolution recurrent neural network (CRNN) and 3D convolution neural network (3DCNN) are used to decode OAM modes with a low bit error rate (BER). Moreover, a graphics processing unit (GPU) is used to accelerate the training process with slight power consumption. The utilized datasets for OAM states are generated by applying different scenarios using a trial-and-error method. The simulation results imply that LDPC-coded VFs achieve the largest peak signal-to-noise ratios (PSNRs) and the lowest BERs using the 16-OAM-SK model. Both 3DCNN and CRNN techniques have nearly the same performance, but this performance deteriorates in the case of larger dataset classes. Moreover, the GPU accelerates the performance by almost 67.6% and 36.9% for the CRNN and 3DCNN techniques, respectively. These two DL techniques are more effective in evaluating the classification accuracy than the other traditional techniques by almost 10 - 20%.

Deep learning (DL) methods have beenachieving amazing results in solving a variety ofproblems in many different fields especially in the areaof big data. With the advances of the big data era inbioinformatics, applying DL techniques, the DNAsequences can be classified with accurate and scalableprediction. The strength of DL methods come from thedevelopment of software and hardware, such asprocessing abilities graphical processing units (GPU) forthe hardware and new learning or inference algorithmsfor the software, which reducing the main primarydifficulties that faced the training process. In This work,we start from the previous classification methods such asalignment methods pointing out the problems, which areface to use these methods.After that, we demonstratedeep learning, from artificial neural networks to hyperparameter tuning, and the most recent state-of-the-artDL architectures used in DNA classification. After that,the paper ended with limitations and suggestions.

Currently, secure multimedia applications are becoming a very hot research topic, specifically over the Internet and wireless communication networks due to their rapid progress.Several researchers have implemented various chaotic image and video encryption algorithms to achieve data stability and communication security.This paper presents a novel bit-level video frame cryptosystem that is dependent on the piecewise linear chaotic maps (PWLCMs).It is implemented for orbital angular momentum (OAM) modulation over different turbulence channels.Firstly, the mathematical model for the bit error rate (BER) of OAM modulation is derived over the gamma-gamma turbulence channel.After that, a comparison between the theoretical results from Mathematica and the simulation results from MATLAB for different turbulence strengths, signal-to-noise ratios (SNRs), and propagation distance values is presented to assure that there is a perfect match.The proposed video cryptosystem is checked using entropy analysis, histogram testing, attack analysis, time analysis, correlation testing, differential analysis, and other quality and security evaluation metrics.The simulation results and the performance analysis confirm that the proposed cryptosystem is reliable and secure for video frame encryption, and communication with different turbulence conditions in free space.

A hybrid multi-state orbital angular momentum-multi pulse-position modulation ( $N$ OAM-MPPM) scheme over gamma-gamma free-space optical ( ${\Gamma \Gamma }$ -FSO) channel is studied in this paper. In our study, all atmospheric and pointing error impacts are taken into account. Expressions for the parameters of ${\Gamma \Gamma }$ -FSO-pointing error channel are derived. In addition, approximate-tight upper bounds on the bit-error rates (BERs) of $N$ OAM and $N$ OAM-MPPM techniques are developed over ${\Gamma \Gamma }$ -FSO-pointing error channels, considering the influences of beam divergence and pointing error (PE). The ${\Gamma \Gamma }$ -FSO-PE channel parameters and the BER expressions are evaluated numerically and verified by simulation. It turned out that the analytical results are nearly the same as those obtained from simulation under different turbulence scenarios and OAM modes. The results demonstrate that under variable turbulence conditions, the $N$ OAM-MPPM technique outperforms both ordinary $N$ OAM and MPPM systems. Furthermore, different deep learning (DL) techniques, namely random forest (RF), convolution neural network (CNN), and auto-encoder (AE), are employed to get the optimum classification accuracy using different datasets of $N$ OAM-MPPM over ${\Gamma \Gamma }$ -PE channel model. Finally, the results indicate that AE has the best performance metrics compared to other models using different datasets.