Smart Computing and Systems Engineering - 2025 (SCSE 2025)
Permanent URI for this collectionhttp://repository.kln.ac.lk/handle/123456789/30037
Browse
32 results
Search Results
Item An Integrated Deep Learning Framework for Early Detection of Vision Disorders(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Jayathilaka, S.; Balaruban, D.; Kumanayake, I.; Elladeniya, A.; Wijendra, D.; Krishara, J.; De Silva, M.Vision impairment due to retinal diseases like Diabetic Retinopathy (DR), Age-Related Macular Degeneration (AMD), Glaucoma, and Retinal Vein Occlusion (RVO) poses a significant health challenge in Sri Lanka, where these conditions are leading causes of blindness. This research presents a novel multi-disease prediction system leveraging advanced deep learning techniques for early detection of DR, AMD, Glaucoma, and RVO. The study utilized publicly available datasets, including retinal fundus images from repositories such as RFMiD, IDRiD, APTOS validated by medical professionals to ensure diagnostic reliability. These images were preprocessed and augmented to train robust convolutional neural network (CNN) models tailored to each disease. The predictive models were developed and optimized using hybrid architectures, integrating attention mechanisms and feature fusion for enhanced performance. This approach achieved high accuracies—93% for DR, 92% for AMD, 94% for Glaucoma, and 94% for RVO—demonstrating robustness and consistency across diverse retinal conditions. To validate real-world applicability, the models underwent further testing in clinical settings using a Sri Lankan dataset, reflecting local disease prevalence and imaging conditions. By combining validated public data with clinical testing, this scalable system supports ophthalmologists in early diagnosis, reducing diagnostic delays and improving patient outcomes. This work offers a reliable, innovative solution to mitigate the burden of blindness in Sri Lanka and beyond.Item A Systematic Literature Review of Deepfake Face Image Detection with Transfer Learning Techniques(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Wimalasena, W.; Herath, H.; Hewapathirana, IArtificial intelligence-based fake content generation, also known as deepfake media content, is spreading rapidly with the advancement of technology. This can manipulate or completely change the real media by threatening the authenticity of the data. With the availability of major amounts of data, deepfake face image generation has become an emerging issue in today’s digitalized world. Therefore, it is crucial to understand the most effective methods to detect this deepfake face image content. This paper provides a comprehensive discussion about how transfer learning can be used in this specific area to detect deepfake face images through a systematic literature review of (49) papers using the PRISMA method. The paper addresses the gap in up-to-date detailed analysis of transfer learning methods in this area which emphasizes the effectiveness of the process of deepfake image detection by studying papers in the past decade. This paper presents a comprehensive discussion of the models that have been used in this area along with the transfer learning approaches that have been used. The key findings highlight that the most common transfer learning approach for deepfake image detection is CNN combined and the most effective model for deepfake image detection in transfer learning is fine-tuned models. This paper also highlights the challenges and possible improvements in the area of deepfake face image detection with transfer learning techniques.Item A Deep Learning-Based Approach for Detecting Duplicate GitHub Issues in Open-Source Repositories Using LSTM(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Dharmadasa, T. K. R. S.; Rupasingha, R. A. H. M.; Kumara, B. T. G. S.GitHub is a platform used along with the popular version control tool Git to provide hosting facilities to software repositories. Users can publish GitHub issues to notify the repository contributors about bugs, questions, and feature requests. GitHub hosts open-source repositories that are contributed by developers across the globe. The asynchronous and uncoordinated nature of these contributions in open-source repositories increases the probability of posting duplicate GitHub issues, resulting in redundant efforts. The standard mechanism introduced by GitHub to mark duplicate issues is adding a comment to that issue body mentioning the original issue. Then GitHub will add the corresponding duplicate tag and close that issue. However, due to manual labor required to find duplicates, developers are discouraged from seeking similar issues before publishing a new issue to GitHub. The study’s main objective is to address this problem and propose an automated solution using deep learning algorithms. Our research introduces a novel approach that combines feature extraction and similarity calculations to identify duplicate GitHub issues. The proposed methodology extracted over 4000 GitHub issues covering different programming languages and repositories. After pre-processing, various features were extracted using multiple feature extraction techniques, and semantic similarity metrics such as cosine similarity were utilized to create the feature vector. The feature vector was used with different algorithms like Artificial Neural Network (ANN) and Recurrent Neural Network (RNN) including deep-learning algorithms like Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). Algorithm results are compared to detect the most suitable approach for detecting duplicate GitHub issues. Based on the different evaluations, LSTM is the better approach resulting in 88% accuracy with the highest precision, recall, and f-measures while giving the lowest error rates. With this proposed methodology, duplicate GitHub issues can be easily detected, reducing the manual work.Item Machine Learning-Based Detection of ARP Spoofing Attacks Using Behavioral Analysis(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Sumanasekara, S. G.; Abeysinghe, D. V. D. S.This research focuses on identifying and reducing ARP (Address Resolution Protocol) spoofing attacks, which pose a significant vulnerability in network security. These attacks allow attackers to manipulate data flows by linking their MAC address with a legitimate IP address. The study aims to develop a robust framework for detecting ARP spoofing behaviors and mitigating potential network attacks. The research first involved performing a behavioral analysis on ARP traffic to extract relevant features, such as ARP request frequency, IP-MAC mapping inconsistencies, time between requests, and other typical network behaviors that indicate spoofing. Various machine learning techniques were then employed, including models like Linear SVC, Logistic Regression, K-Nearest Neighbors (KNN), and Gaussian Naïve Bayes. Among these models, KNN achieved the highest accuracy of 0.94, demonstrating its effectiveness in identifying spoofing behaviors. The overall performance of the framework highlights the potential of combining behavioral analysis with machine learning to enhance network security by detecting and mitigating ARP spoofing attacks in real-time.Item Leveraging Large Language Models for Addressing Operational Challenges in Sri Lankan SMEs(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Amarakoon, M.; Rajapakse, C.Small and Medium Enterprises (SMEs) are critical to Sri Lanka's economy but face challenges like regulatory complexity, limited technology access, and linguistic barriers. This research explores the use of Large Language Models (LLMs) to address these issues by localizing AI for Sinhala and Tamil. The study fine-tunes the Aya Model, a multilingual instruction-tuned LLM, using Sri Lankan business and regulatory datasets. Techniques such as data augmentation and cross-lingual transfer learning were applied to mitigate the scarcity of structured Sinhala data, ensuring cultural and linguistic alignment. Evaluation shows a 30% improvement in regulatory compliance accuracy, faster response times for customer queries, and enhanced operational efficiency. The localized LLM automates multilingual customer interactions and generates context-sensitive business insights, offering a scalable, cost-effective solution for SMEs. This study demonstrates how LLMs can bridge technological gaps, enabling Sri Lankan SMEs to achieve sustainable growth and competitiveness in a resource-constrained setting.Item Hybrid Deep Learning for Portable Weather Forecasting: Real-Time Predictions Using CNN- LSTM-Transformer Models(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Perera, K. P. V. D. U. P.; Fernando, K. J. P.; Lakmak, H. K. I. S.; Nirmal, W. C.Accurate weather forecasting is critical for agriculture, disaster management, and transportation sectors. However, traditional forecasting systems often require extensive computational resources and centralized infrastructure, limiting their accessibility in remote and underserved regions. This study introduces a Portable Weather Forecasting Station that combines real-time sensor-based data acquisition with advanced deep learning techniques. The station integrates sensors to measure parameters such as temperature, humidity, wind speed, and solar radiation, processed by a Raspberry Pi for localized predictions. A hybrid deep learning model comprising Convolutional Neural Networks (CNNs), Long Short-Term Memory Networks (LSTMs), and Transformers is developed to capture both short-term patterns and long-term dependencies in the data. The system's performance is enhanced through hyperparameter optimization using Optuna, with metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Skill Score used for evaluation. The hybrid model demonstrated superior accuracy compared to standalone architectures. Designed for autonomy with battery backup, the station operates independently of external infrastructure, making it ideal for deployment in resource-constrained environments. This research offers an innovative approach to localized, real-time weather forecasting, addressing the limitations of traditional methods while ensuring accessibility and scalability.Item Deep Learning based Screen Display Fault Detection System for Vehicle Infotainment Applications(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Ramesh, B.; Dheeba, J.; Raja Singh, R.Modern vehicles are integrated with in-vehicle infotainment systems and are subject to software faults. This paper explores the application of deep learning algorithms to identify visual defects in infotainment systems and automatically document the issues. A real-time capable framework is deployed, delivering immediate feedback on detected defects. The proposed system performs thorough analysis, automatically summarizes detected defects, and generates detailed reports, significantly reducing manual documentation effort and supporting faster decision-making. The performance of the developed models is evaluated using Convolutional Neural Networks (CNN) and Artificial Neural Network (ANN) classifiers. Experimental results demonstrate the superior performance of the CNN model, achieving a training accuracy of 82.21% with an F1 score of 0.85, and a testing accuracy of 80.51% with an F1 score of 0.811. In comparison, the ANN model achieves a training accuracy of 70.18% with an F1 score of 0.7314, and a testing accuracy of 69.32% with an F1 score of 0.705.Item FITSTYLE: An Application Revolutionizing Online Shopping by Enhancing the Virtual Try-On Experience(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Divyanjalee, G.; Ilmini, K.; Uwanthika, I.In the last decade, the fashion industry has been significantly influenced by the rise of e-commerce and mobile commerce. As online consumers, we often face the challenge of selecting the right clothing size without the ability to physically try on garments, leading to frustration, uncertainty, and high return rates. These issues negatively impact customer satisfaction and online sales productivity, highlighting the need for advanced virtual shopping solutions. To address this problem, FITSTYLE proposes a sophisticated virtual try-on tool based on Generative Adversarial Networks (GANs) to simulate how clothes fit a consumer’s body. The FITSTYLE system tackles challenges in online clothing shopping by utilizing advanced technologies such as ResNet101 for image preprocessing, OpenPose for pose estimation, image segmentation to isolate users from backgrounds, and garment deformation algorithms for accurate fitting. Designed for ease of use and realistic visualization, it enhances customer satisfaction and reduces return rates. The research employed a mixed quantitative and qualitative methodology, collecting data from online shoppers to understand user needs and preferences. Based on these insights, the system was implemented using the most suitable technologies, with initial results showing improved customer decision-making and engagement. The system enhances satisfaction, reduces return rates, and boosts productivity in online sales, while also paving the way for future advancements in virtual try-on technologies and the broader fashion e-commerce landscape.Item Computer Vision Technique for Quality Grading of Cardamom Spice(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Ahnaf, M. R. A.; Nizzad, A. R. M.; Mafaza, N. F.In the agriculture industry, spices such as cardamom must be graded and their quality evaluated to maintain product standards and increase market value. Traditionally, cardamoms are graded manually, but this process is time-consuming, inconsistent, and prone to human error due to fatigue and subjectivity. This study proposes computer vision-based techniques to detect and categorize cardamom into four grades—premium, standard, substandard, and defective—based on size, shape, and texture. Researchers prepared 880 images representing these categories under uniform lighting conditions. Preprocessing techniques, including resizing, noise addition, and augmentation, were applied to enhance the dataset for robust model training. YOLOv8 was utilized for the detection and grading of cardamoms due to its capability for real-time object detection and classification. The proposed method effectively differentiated between quality grades with an accuracy of 92%, demonstrating reliability and efficiency. A user-friendly interface was developed using the Streamlit library, allowing users to upload images and obtain grading results instantly. This system offers a practical and scalable solution for improving quality control processes in the agricultural sector. Future work aims to incorporate larger datasets, texture analysis, and integration with automated machinery to further enhance its applicability in industrial settings. The proposed solution can also be generalized to other use cases in the agricultural industry.Item AI-Driven Approach for Measuring and Classifying Diabetic Retinopathy Severity(Department of Industrial Management, Faculty of Science, University of Kelaniya., 2025) Shaik, A.; Little Flower, K.; Veerabhadraiah, S.; Nandini, A.; Sai Kiran, C.; Yashwanth Goud, K.Diabetic Retinopathy is one of the most common complications affecting people with diabetes and is a leading cause of blindness worldwide. Advanced technological methods through image analysis and artificial neural networks have become major assets in addressing the escalating problem of DR. This paper discusses various approaches to implementing automation in DR detection, focusing on image acquisition and preprocessing, feature extraction, and classification using AI. We review the utility of these systems in terms of cost, benefits, and performance, as well as the challenges related to data quality, model interpretability, and regulatory requirements. The study demonstrates that automation holds the key to delivering higher patient-impact opportunities in clinical applications such as screening programs and telemedicine. Finally, we discuss directions for future research and community implementation, emphasizing the importance of highly controlled, naturalistic designs and the translation of research findings into clinical practice.