Title– A MULTI-SOURCE CBIR APPROACH LEVERAGING AUTOENCODERS FOR IMAGE ORGANIZATION AND DEEP LEARNING FOR SURROGATE LABELING

Author– Manish Rai and Rupali Sharma

Abstract

Deep learning-based Content-Based Image Retrieval (CBIR) has emerged as a leading research area due to its ability to deliver more accurate search results than traditional methods, despite being computationally demanding. This study presents an innovative approach that enhances CBIR performance through a multilevel aggregation technique integrated with autoencoders, aimed at precise feature selection and improved search accuracy. The proposed method employs a dual strategy. First, a multilevel aggregation technique is used to process image features at various granularities, capturing both local and global image attributes for a comprehensive representation. Autoencoders then reduce the dimensionality of these features, retaining only the most relevant information while reducing computational costs. The refined features are leveraged to effectively tag the search key, guiding the search process toward the most pertinent target images. Second, the method distinguishes between locally significant datasets and generic ones, employing specific approaches for each. For domain-specific datasets, unique characteristics are utilized to enhance retrieval rates, while a more generalized strategy is applied to broader datasets. Query expansion is implemented to broaden the search scope, incorporating additional relevant terms or images related to the original query. Additionally, pseudo-labeling is introduced, where deep learning models classify images into positive (similar to the query) and negative classes. Query image features are compared to those in the search pool using assigned weights, and target images are ranked based on an adaptive threshold. Tested on public datasets, this method demonstrates significant improvements in precision, recall, and computational efficiency compared to recent approaches. This robust technique shows promise for applications requiring precise image retrieval, such as medical imaging, security surveillance, and digital asset management.

Index Term– Deep learning; CBIR; review; image retrieval

DOI- 10.30696/IJEEA.XII.III.2024.01-18

Reference to this paper should be made as follows:  Manish Rai and Rupali Sharma, (2024), “A Multi-Source CBIR Approach Leveraging Autoencoders for Image Organization and Deep Learning for Surrogate Labeling” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 1-18.

Title– A MODIFIED XCEPTION FRAMEWORK WITH ATTENTION LAYERS FOR RECOGNIZING AIR LEAKS IN THE LUNGS

Author– Alex M. Goh and Xiaoyu L. Yann

Abstract

Chest radiographs play a vital role in diagnosing various lung disorders, particularly pneumothorax, which poses significant health risks if not detected promptly. This study investigates the application of artificial intelligence (AI) to enhance the classification accuracy of pneumothorax from chest X-ray images. By integrating the Xception network with an attention module, researchers developed a robust model tested on a dataset comprising 2,597 chest X-ray images. The model exhibited exceptional performance metrics, achieving a training accuracy of 99.18% and a validation accuracy of 87.53%. Additionally, it attained an impressive average area under the ROC curve (AUC) of 90.00%. These results highlight the potential of AI technologies in significantly improving diagnostic accuracy for pneumothorax and other pulmonary conditions. The incorporation of attention mechanisms allows the model to focus on the most informative regions within the X-ray images, which is crucial for detecting subtle indicators of pneumothorax that may be overlooked by traditional methods. Such advancements in AI-driven diagnostic tools could lead to earlier intervention and better patient outcomes. The findings of this study not only underscore the efficacy of AI in radiological assessments but also pave the way for further exploration into automated diagnostics across various medical imaging domains, potentially transforming clinical practices in lung disease detection.

Index Term– Pneumothorax; Xception; Deep Learning; Channel Attention; Spatial Attention; Transfer Learning.

DOI- 10.30696/IJEEA.XII.III.2024.19-32

Reference to this paper should be made as follows:  Alex M. Goh and Xiaoyu L. Yann, (2024), “A Modified Xception Framework with Attention Layers for Recognizing Air Leaks in The Lungs” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 19-32.

Title– UNRAVELING THE SPREAD OF FALSE INFORMATION AMONG INDIAN COLLEGE STUDENTS DURING COVID-19: AN ANALYTICAL APPROACH

Author– Rakesh Kumar Pandey and Sabina Raj Priyadarshi

Abstract

In recent years, social media has become an increasingly popular tool for sharing information, particularly among college students in India. However, this rise has also led to an increase in the spread of misinformation. This study explores the motivations behind why college students share false information online. Using a survey conducted via Google Forms, data was collected on demographics, motivations, and the characteristics of information shared. A Misinformation Sharing Index (MSI) was developed to analyse patterns. The study found that 62.4% of students admitted their friend’s shared misinformation, primarily driven by factors like self-expression, socializing, and information characteristics. Students who accessed social media more than 12 times daily were most likely to share misinformation. Interestingly, verifying the accuracy of information was not a top priority for many. Gender differences were also observed, with male students sharing more misinformation than females, although female respondents believed their peers shared more. An educational gap emerged, showing that undergraduate students, especially males, shared more misinformation than their counterparts. The COVID-19 pandemic amplified the spread of false information, leading to unnecessary lifestyle changes and negatively affecting mental health. The research highlights that misinformation sharing has grown significantly during the pandemic, resulting in harmful practices and behavioural changes. This study offers unique insights into the underlying reasons for misinformation sharing among college students and sheds light on the need for greater awareness regarding information verification on social media platforms.

Index Term- COVID-19; Misinformation; social media; College students; Mental health.

DOI- 10.30696/IJEEA.XII.III.2024.33-50

Reference to this paper should be made as follows:  Rakesh Kumar Pandey and Sabina Raj Priyadarshi, (2024), “Unraveling The Spread of False Information Among Indian College Students During Covid-19: An Analytical Approach” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 33-50.

Title– INVESTIGATING ECO-FRIENDLY RENDERING SOLUTIONS FOR 3D MODELS ON MULTI-CORE PROCESSORS

Author– V.V.S. Rao, Mukesh Ranjan and Prof. V. Kunwar

Abstract

Graphics Processing Units (GPUs) are vital for handling computationally intensive tasks, particularly in real-time 3D applications. Designed with additional processing power, GPUs are optimized to efficiently render complex scenes. Shaders, which are small programs running on the GPU, control how each pixel is processed and rendered. They affect key aspects of the rendering pipeline, such as geometric complexity, texture resolution, and per-pixel lighting. However, the extensive use of shaders can lead to increased power consumption, especially when managing complex 3D scenes. This study investigates GPU power usage during the rendering process and proposes a dynamic power prediction model. Initially, the power consumption of various rendering configurations is analyzed to understand the impact of shader parameters. Based on this analysis, a power-aware, dynamically reconfigurable rendering model is introduced, designed to optimize power consumption through shader calls while maintaining performance and visual quality. The implementation integrates TensorFlow management libraries and dynamic voltage and frequency scaling (DVFS), a technique that adjusts the GPU’s voltage and frequency based on workload demands. Shader parameters such as geometric complexity, texture resolution, per-pixel lighting, and filtering are evaluated for their power usage during common GPU tasks. By continuously monitoring the framerate, the system dynamically adjusts the GPU frequency to optimize power efficiency. A lower framerate limit of 30 frames per second (fps) and an upper limit of 60 fps are established. If the framerate exceeds 60 fps, the GPU frequency is reduced to conserve power, while the frequency is increased if the framerate drops below 30 fps to maintain smooth performance. This dynamic adjustment ensures the GPU operates efficiently without sacrificing the quality of 3D rendering. The experimental results demonstrate that the proposed model offers around 40% power savings compared to previous methods. Additionally, it improves computational speed while preserving the quality of the rendered scenes. This approach highlights the potential for achieving a balance between power efficiency and rendering performance in GPUs, making it ideal for power-sensitive applications.

Index Term- GPU; 3D graphics rendering; Shader parameters; TensorFlow; DVFS; Workloads.

DOI- 10.30696/IJEEA.XII.III.2024.51-68

Reference to this paper should be made as follows:  V.V.S. Rao, Mukesh Ranjan and Prof. V. Kunwar, (2024), “Investigating Eco-Friendly Rendering Solutions For 3d Models on Multi-Core Processors” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 51-68.

Title– SMART VIRTUAL ENVIRONMENT FOR DESIGNERS WITH MOTOR LIMITATIONS

Author– Dr. N. Muni Nordin

Abstract

Traditional CAD modeling software relies heavily on input devices such as keyboards and mice for the creation of three-dimensional (3D) models, which can be challenging for individuals with motor disabilities. These tasks require fine motor control, which may be difficult for users who experience limited mobility in their limbs—especially their arms and fingers—due to injury or congenital conditions. To address these accessibility barriers, this paper presents a virtual reality (VR)-based interface that enables users with motor disabilities to create simple 3D architectural models. The proposed system leverages head-gaze tracking as a primary input method in the VR environment, allowing users to manipulate and scale 3D objects, such as simplified building models, without needing to use their hands. Additionally, the VR system incorporates head tilting for navigation, with the user seated in a revolving chair, eliminating the need for any limb movement. This approach minimizes physical effort and improves accessibility for individuals with limited motor abilities. Developed using the Unity game engine, the system offers two variations of the VR interface, each with a different button layout for constructing cuboidal volumes that represent architectural structures. The two versions were tested with 32 participants, with performance assessed based on task completion time. In addition to these objective measures, participants also provided feedback on perceived effort and visual clutter. Following these tests, retrospective feedback sessions further informed the evaluation of each interface variant. The results indicated that Variant 1 performed better overall, showing higher usability and lower effort requirements. In addition, the paper proposes a framework for incorporating artificial intelligence (AI) into the VR interface, creating an intelligent and adaptive system that can adjust in real time to users’ capabilities and challenges. This AI-driven approach could lead to a more robust, gaze-based VR tool with broader functionality, enabling individuals with motor disabilities to undertake more complex 3D modeling tasks in the future.

Index Term- Virtual Reality; design tool; intelligent interface; gaze-based input; artificial intelligence.

DOI- 10.30696/IJEEA.XII.III.2024.69-82

Reference to this paper should be made as follows:  Dr. N. Muni Nordin, (2024), “Smart Virtual Environment for Designers with Motor Limitations” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 69-82.

Title– AN EFFICIENT SKIN DISEASE DETECTION APPROACH USING OPTIMIZED REGION GROWING SEGMENTATION AND AUTOENCODER-BASED CLASSIFICATION

Author– Mukesh Kumar Prabhakar and Prof. S K Singh

Abstract

Detecting skin disorders through visual examination is difficult due to the complex nature of lesions, background skin textures, hair, and varying lighting conditions. While computer vision and machine learning have made progress in diagnosing skin lesions, existing methods face limitations in handling these complexities. To address this, we propose a novel detection framework aimed at improving the accuracy of skin disease diagnosis. Our approach begins with segmenting the diseased lesion using Optimized Region Growing, enhanced by Grey Wolf Optimization (GWO), which effectively isolates lesions from complex backgrounds. Once segmented, texture features are extracted using the Gray Level Co-occurrence Matrix (GLCM) and Weber Local Descriptor (WLD). GLCM captures key texture characteristics such as contrast and homogeneity, while WLD focuses on local intensity variations, providing robustness against artifacts like skin texture variations and illumination issues. Next, an autoencoder is utilized to reduce the dimensionality of the extracted features by generating a compact latent representation. This helps preserve essential information while improving the model’s ability to differentiate between diseased and healthy lesions. Finally, a convolutional neural network (CNN) is used to classify the lesions based on the latent features. By integrating GWO for segmentation, advanced texture feature extraction, and autoencoder-based classification, the framework achieves superior diagnostic performance. Experimental results show that this method outperforms traditional deep learning strategies, particularly in addressing the complex and overlapping structures of skin lesions. This makes it a promising tool for more precise and automated skin disease detection.

Index Term- Grey Wolf Optimization; Gray Level Co-occurrence Matrix; Weber Local Descriptor.

DOI- 10.30696/IJEEA.XII.III.2024.83-95

Reference to this paper should be made as follows:  Mukesh Kumar Prabhakar and Prof. S K Singh, (2024), “An Efficient Skin Disease Detection Approach Using Optimized Region Growing Segmentation and Autoencoder-Based Classification” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 83-95.

Title– IMPROVING ANOMALY DETECTION IN INTRUSION SYSTEMS USING GROUP APPROACHES

Author– Syafrika Deni Minz and Iskandar Joe

Abstract

The rapid expansion of IoT-based applications across various sectors, particularly in Smart Cities, has brought significant benefits but also heightened security concerns. As IoT infrastructure acts as the core system enabling the seamless operation of these smart environments, it faces an increased risk of cyberattacks. The continuous communication between IoT devices and cloud services through embedded sensors creates opportunities for malicious actors to exploit vulnerabilities in these channels, making them susceptible to attacks. This paper proposes a sophisticated anomaly detection method to bolster the cybersecurity of Smart Cities by enhancing intrusion detection systems (IDS). The proposed method leverages multiple machine learning techniques, including Support Vector Machines (SVM), Artificial Neural Networks (ANN), k-Nearest Neighbor (KNN), Linear Regression (LR), Decision Trees (DT), and Random Forest (RF). These algorithms are utilized to detect anomalies in IoT networks, helping to identify and mitigate potential cyber threats. A key aspect of this approach is the use of ensemble techniques, such as bagging and boosting, which provide an additional layer of security to the detection architecture. Ensemble methods improve model performance by combining the strengths of various classifiers, thereby offering more accurate and reliable threat detection. The paper moves beyond traditional approaches that rely on single classifiers and emphasizes the integration of cross-validation and feature selection to optimize detection outcomes. The proposed model is evaluated using two widely recognized datasets, UNSW-BC15 and CICIDS2017. The experimental results, measured in terms of Accuracy, Precision, Recall, and F1 Score, show that this ensemble-based approach outperforms existing state-of-the-art techniques. It proves particularly effective in detecting rare and sophisticated attacks, offering a robust and scalable solution to secure IoT infrastructures in Smart Cities against evolving cyber threats.

Index Term- IoT, smart city, SVM, decision tree, KNN, linear Regression, ANN, cybersecurity, bagging, boosting, intrusion detection system, ensemble techniques.

DOI- 10.30696/IJEEA.XII.III.2024.96-109

Reference to this paper should be made as follows:  Syafrika Deni Minz and Iskandar Joe, (2024), “Improving Anomaly Detection in Intrusion Systems Using Group Approaches” Int. J. Electronics Engineering and Applications, Vol. 12, Issue 3, pp. 96-109.

Title– BRIDGING THE GAP BETWEEN AI AND QUANTUM COMPUTING

Author– Rajesh Kumar Tiwari, Dr. Abu Bakar Bin Abdul Hamid and Dr. Tadiwa Elisha Nyamasvisva

Abstract

This study investigates the convergence of machine learning and quantum computing, with a particular emphasis on the ways in which machine learning techniques can enhance quantum algorithms. The primary focus areas include quantum data analysis, quantum machine learning, and hybrid quantum-classical approaches, illustrating how these methodologies are bridging the divide between artificial intelligence (AI) and quantum computing. The research examines the potential of quantum data generation to facilitate machine learning applications, as well as the implementation of quantum-assisted optimization to address complex problems more efficiently than conventional methods. Furthermore, it explores the development of quantum neural networks, which leverage quantum computing capabilities such as superposition and entanglement, potentially leading to significant advancements in AI. By analyzing these trends, the study offers valuable insights into how the integration of quantum computing with AI could transform data processing, problem-solving, and model training, resulting in innovative applications and enhanced computational efficiency across diverse fields. Reason: The revised text maintains the original content while enhancing clarity, coherence, and academic tone.

Index Term- Machine Learning, Quantum Computing, Artificial Intelligence, Quantum Algorithms, Hybrid Approaches.

DOI- 10.30696/IJEEA.XII.III.2024.110-125

Reference to this paper should be made as follows:  Rajesh Kumar Tiwari, Dr. Abu Bakar Bin Abdul Hamid and Dr. Tadiwa Elisha Nyamasvisva, (2024), “BRIDGING THE GAP BETWEEN AI AND QUANTUM COMPUTING: MACHINE LEARNING FOR QUANTUM ALGORITHMS” Int. J. of Electronics Engineering and Applications, Vol. 12, No. 3, pp. 110-125, DOI 10.30696/IJEEA.XII.I.2024.110-125.