ASPG Menu
search

American Scientific Publishing Group

Research Feed

Found 3720 matches for "All Articles"

SCNN-UNet: A Novel Deep Learning Approach for Pulmonary Embolism Detection in COVID-19 Patients Using Super Pixel Segmentation

Inventory management is crucial for optimizing consumer demand and supply chains in e-commerce companies.  This is the stage at which precise inventory forecasting becomes necessary for forecasting future demand patterns and stock levels.  Traditional forecasting methods often struggle with e-commerce data due to seasonality, sudden changes in customer behavior, and nonlinearity.  Machine learning (ML) and deep learning (DL) techniques have become powerful weapons for inventory prediction because they can analyze huge amounts of data with high dimensionality. E-commerce firms can improve their resource allocation, inventory management, and customer experience in highly competitive market environments.  This paper proposes different types of inventory forecasting models and mainly evaluates the applicability of sophisticated machine learning algorithms.  While we commonly use old methods like Random Forest, ARIMA, and MLPs, they often lack the necessary robustness to nonlinearity within inventory data.  To address these problems, we introduce a novel method that combines convolutional neural networks (CNN) and XGBoost called CNN-XGBoost, which provides better feature extraction than the conventional prediction model and regression performance.  We then compared CNN-XGBoost's performance to traditional forecasting methods (another common approach to contextualizing predictive model performance) using a 52-week simulated dataset in which we mimic patient data growing over time.  We used key performance metrics such as R2, mean squared error (MSE), and mean absolute percentage error (MAPE) to assess each model's accuracy.  The CNN-XGBoost model performed much better than others, with an R2 of 0.78, which means our proposed model can explain more variation compared to other competitors, as depicted in the results section.  It also had the best MSE of 0.15, indicating better predictive performance.  The CNN-XGBoost model demonstrated promising prospects as a robust inventory forecasting tool for commerce despite its slightly higher MAPE value (0.69), suggesting some vulnerability to outlier data points.  This study demonstrates the potential of using a convolutional neural network in combination with gradient boosting techniques to tackle the complexity of stock management issues and the fact that it outperforms based line methods by a large margin.

groups
Sukhwinder Bir mail -
Vijay Dhir mail
link https://doi.org/10.54216/JISIoT.160214

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Alzheimer Detection Using Deep Learning Methods

This study proposes a deep learning-based framework to detect and classify Alzheimer's disease (AD) in the early stages using medical imaging, and specifically Magnetic Resonance Imaging (MRI). Specifically, we propose a Convolution Neural Network (CNN) based model and transfer learn (MobileNet) through pre-trained models based on task domain to improve model performance on binary AD classification. Thanks to minimizing computational complexity and memory costs, the model with 99.86% accuracy rate can mitigate overfitting and is an ideal approach for real time and eco-friendly monitoring of AD evolution. The findings suggest that the model could help clinicians in diagnosing AD even based on MRI images, which has great potential as a scalable and efficient solution for the early-stage diagnosis and classification of the disease. Our work will include the addition of further pre-trained models, increased dataset size via data augmentation, and the application of MRI segmentation to better isolate some of the key features of Alzheimer.

groups
Raghad K. Mohammed mail -
Mohammed Q. Jawad mail -
Othman Mohammed Jasim mail
link https://doi.org/10.54216/JISIoT.160215

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Enhancing Osteoporosis Detection with Fuzzy Logic Preprocing and Pre-Trained Deep Convolutional Neural Networks

This study investigates combining fuzzy logic with deep learning methodologies in classifying X-ray images for osteoporosis detection. Osteoporosis, defined by compromised bone integrity and heightened fracture susceptibility, requires prompt and precise diagnosis for effective treatment. We devised a hybrid approach that amalgamates transfer learning from Convolutional Neural Network (CNN) architectures, including MobileNetV2, AlexNet, ResNet50V2, and Xception, utilizing fuzzy logic during the preprocessing phase to address uncertainty and imprecision in X-ray images, thereby enhancing the quality of the input data for the subsequent pre-trained models. The research entailed the examination of a significant dataset of X-ray images and the implementation of the proposed methodology to categorize images as osteoporotic or non-osteoporotic, attaining a remarkable accuracy of 99.68% and a receiver operating characteristic (ROC) of 100% through the integration of fuzzy logic preprocessing with ResNet50V2. This innovative method may substantially decrease diagnostic inaccuracies and enhance patient outcomes, facilitating additional research and development in applying deep learning techniques in healthcare.

groups
Woud Majid Abed mail -
Murtadha M. Hamad mail -
Azmi Tawfeq Hussein Alrawi mail
link https://doi.org/10.54216/JISIoT.160216

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

A Hybrid Encryption Model with Blockchain Integration for Secure Cloud Data Storage and Retrieval

  Data security, privacy, sensitivity, and integrity are major concerns when using cloud-based storage solutions. In this paper, we propose a hybrid encryption model that has been integrated with blockchain technology to secure data storage in the cloud. The proposed model facilitates data encryption using a symmetric cryptography algorithm for efficient large data encryption while ensuring the encryption key can only be exchanged using asymmetric cryptography. This model utilizes the power of blockchain to manage metadata securely and associated encryption keys to ensure that records are tamper-proof, removing the need for third parties to be trusted. The security, key management, and data integrity of the proposed model are better than traditional cloud storage and existing blockchain-based approaches. The performance evaluation suggests that the model achieves a balance between security and cost efficiency, while moderate transaction speed will be witnessed owing to blockchain operations. Our proposed work aims to provide a scalable, fast, reliable, and decentralized architecture-based solution to address the challenges of cloud data security.  

groups
Firas Mohammed Khalaf mail -
Ali Makki Sagheer mail
link https://doi.org/10.54216/JISIoT.160217

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Efficient Plant Disease Detection Using Lightweight Deep Learning Model

Early detection of plant diseases is critical to minimizing their adverse effects on agricultural productivity. In particular, machine vision and deep learning approaches (e.g., convolutional neural networks, CNNs) have been increasingly applied for automatic plant disease identification. Although existing deep learning models achieve satisfying classification accuracy, they often consist of millions of parameters that significantly lead to the lengthy training time, prohibitive calculation costs and deployment obstacles at the resource-constrained edge devices. In order to overcome those constraints, we introduce a new deep learning architecture, which uses adaptations of Inception layers and residual connections that can help both with feature extraction and efficiency. In addition, depthwise separable convolutions are used to drastically reduce the amount of trainable parameters with small loss of representational power. We perform training and evaluation of the proposed model on three located benchmark plant disease datasets, PlantVillage dataset, the Rice Disease dataset. Experimental results show that our model achieves state-of-the-art classification accuracy of 99.39% on the PlantVillage dataset, 98.66% on the Rice Disease dataset. In contrast to the state-of-the-art deep learning models, our method obtains higher accuracy with fewer parameters so that it could be better suited for real-time applications on mobile and embedded systems. We explore an application of deep learning with the use of optimized architectures and present the viability of this technique in precision agriculture for faster and more accurate diagnosis of diseases in plants with lower computational load.

groups
Abdalrahman Fatikhan Ataalla mail -
Karam Hatem Alkhater mail -
Qusay Hatem Alsultan mail -
Zaid Sami Mohsen mail -
Munther Naif Thiyab mail -
Mohammed Waheeb Hamad mail -
Ahmed Jumaah Yas mail
link https://doi.org/10.54216/JISIoT.160218

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Effective Signal Transmission from Underwater to Air Utilizing Hybrid Communication Systems

Underwater optical communication (UOC) and off-surface areas wireless communications are a rapidly growing field, especially with the emergence of new technologies such as autonomous underwater vehicles and above/water drones. The challenge lies in the absence of a water surface platform to transfer the signal from underwater to off surface. This research investigates the design and implementation of a hybrid communication system that successfully transmits signals from underwater environments to above-water. The study utilizes OFDM as method to generate data on the integration of underwater optical wireless communication (UWOC) at 532nm and LOS optical channel. After adjusting the line of sight through the angle of refraction and overcoming the challenges of water and above water conditions as well as ambient lighting, ambitious results were obtained 100 meters above clear water and 40 meters in haze wither at a depth of 10 meters for transmission. The research has mitigated challenges and enhancing the effectiveness of underwater-to-air communication systems.

groups
Satea H. Alnajjar mail -
Amjed Razzaq Alabbas mail -
Mahmood J. Ahmad mail
link https://doi.org/10.54216/JISIoT.160219

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Improving the Reliability of Wireless Sensor Network Assisted IoT Network with a Cluster-Based Chain-Tree Routing Protocol

The primary objective of designing routing protocols for Wireless Sensor Networks (WSNs) is to extend the network lifetime by optimizing the use of the limited battery energy of the sensor nodes. To improve conservation of energy and longevity of the network in WSNs, this study proposes a Cluster-based Chain-Tree Routing Protocol (CCTRP). Integrating tree based chain and cluster routing methods in WSNs is the primary objective of this study. This new CCTRP adopts a sector-based vertical network-partitioning scheme that divides network into sectors and it again vertically partitions the nodes too form various size of clusters. Then, Minimum Spanning Tree (MST) is created based on the kruskal’s Algorithm through a Chain Leader (CL) node serving as the receiver and chain is formed from CLs of last level cluster to Base Station (BS) in each sector. Using the BS, remaining energy and distance to the next CL node, CCTRP determines the Cluster Leader (CL) or Chain CL node in each cluster. For data transport, it also selects the shortest paths. When the energy that remains in the node is ready to be exhausted, the transition is executed according to this protocol. This results in a significant improvement of the average network lifespan. Finally, the CCTRP protocol outperforms the current protocols in terms of network performance, according to the simulation results.

groups
R. Lalitha mail -
A. V. Senthil Kumar mail
link https://doi.org/10.54216/JISIoT.160220

Volume & Issue

Vol. Volume 16 / Iss. Issue 2

Details open_in_new

Adversarially Robust 1D-CNN for Malicious Traffic Detection in Network Security Applications

While threats in cyberspace are in a state of constant evolution, the use of AI in cyber defense has numerous opportunities and dangers. This paper evaluates adversarial robustness for deep learning networks in network security applications by introducing a novel one-dimensional CNN model for malicious traffic detection. We conducted rigorous end-to-end processing and analysis of network traffic data, using a balanced dataset of 200,000 connections (46.52% benign, 53.48% malicious). Our model architecture includes three convolutional blocks (32, 64, and 128 filters, respectively) with batch normalization and dropout mechanisms (0.3 and 0.2, respectively). We use standardized feature scaling, label encoding for categorical features, and stratified sampling to maintain class distribution integrity.  Our proposed approach achieved remarkable performance metrics compared to standard approaches with a 95% AUC-ROC result (15% better than baseline CNN models) and detection rate of 99.99% malicious traffic (compared to 98.5% with standard architectures). The model demonstrates better robustness with only 10 false negatives out of 107,895 malicious samples, a 67% enhancement compared to current state-of-the-art systems. Training dynamics show great stability with minimal overfitting (validation/training loss difference of only 0.01), indicating good generalization ability.

groups
Baraa Mohammed Hassn mail -
Esraa Saleh Alomari mail -
Jaafar Sadiq Alrubaye mail -
Oday Ali Hassen mail
link https://doi.org/10.54216/JCIM.160113

Volume & Issue

Vol. Volume 16 / Iss. Issue 1

Details open_in_new

Modify Block Chain Environment based on Post-quantum Algorithms

Blockchain technology provides reliable data storage and secures transactions, however, is not suitable for devices with low resources because of its high computational and resource requirements. As quantum computing develops, it poses concerns regarding a cryptographic integrity of blockchain, making them more vulnerable to attacks. Blockchain technology is being used to enhance security and performance. The application of the post-quantum Ascon algorithm in a blockchain setting is presented in this paper. The Ascon hashing algorithm offers a lightweight, efficient architecture for resource-constrained applications, including mobile devices or Internet of Things-based blockchains. By providing high-speed hashing, authentication features, and defense against quantum attacks, it enhances performance and guarantees strong security without putting a strain on network infrastructure. The experimental results show using the Ascon algorithm in a blockchain environment is successful in reducing resource usage and execution time and significantly increasing randomness and unpredictability. Post-quantum Ascon algorithms overcome the drawbacks of traditional technologies and ensure that blockchain systems continue to withstand the new risks posed by quantum computing while increasing overall efficiency

groups
Rasha Hani Salman mail -
Hala Bahjat Abdul Wahab mail
link https://doi.org/10.54216/JCIM.160112

Volume & Issue

Vol. Volume 16 / Iss. Issue 1

Details open_in_new

Computer Vision of Smile Detection Based on Machine and Deep Learning Approach

Smile detection and recognition have been a key component of sentiment analysis, social robotics, human-computer interaction, and mental health monitoring before the advent of deep learning. Understanding and accurately identifying smiles can provide deep insights into human behavior, strengthen communication systems, and enhance adaptive responses in AI interfaces. This paper is a comprehensive review of algorithms developed for smile detection and recognition, and categorizes their main approaches into three traditional computer vision techniques: feature-based, machine learning-based, and deep learning-based. These techniques rely on handcrafted features such as edges, geometric features of the face, and texture, which give interpretability and limited adaptability. This paper explores feature extraction methods such as geometric and histogram-based features (e.g., histograms of directed gradients). In addition, this paper evaluates the effectiveness of traditional classifiers, including support vector machines that use machine learning-based methods, leveraging algorithms such as support vector machines (SVMs), extracted features to classify smiles with improved accuracy. Deep learning techniques, especially convolutional neural networks (CNNs) and hybrid methods provide end-to-end learning capabilities, extracting features directly from raw pixel data and enabling real-time performance. These frameworks, including recurrent neural networks (RNNs) for temporal analysis, generative adversarial networks (GANs) for data augmentation, and graph neural networks (GNNs) for structural analysis, have also pushed the boundaries of smile detection in dynamic and challenging environments. It also aims to provide a comprehensive overview of these classical methods, and analyze their strengths, limitations, drawbacks, and performance across diverse datasets of the proposed databases by focusing on describing these datasets and researchers’ methods of working on them as benchmarks for their research, and highlighting their importance in the environments and their contributions to the development of smile detection algorithms in the field of computer vision. Among these datasets are datasets such as CK+, FER2013, AffectNet, and Jaffe in developing, training, and evaluating smile detection and recognition algorithm models. By comparing these methodologies, our paper recommends directing future research towards more efficient, robust, and scalable solutions for smile detection and recognition in diverse applications.

groups
Huda Lafta Majeed mail -
Oday Ali Hassen mail -
Dhyeauldeen A. Farhan mail -
Yu Yu Gromov mail -
Kavita Sheoran mail -
Geetika Dhand mail
link https://doi.org/10.54216/JCIM.160115

Volume & Issue

Vol. Volume 16 / Iss. Issue 1

Details open_in_new