Recent Submissions
Development of a Sensory Vest for Hearingimpaired (Deaf) Individuals, Focusing on Sensory Substitution Through Haptic Feedback
(Institute of Electrical and Electronics Engineers Inc., 2026-02-05) Ranawana S.R.W.M.A.G; Malasinghe, L
Deafness and hearing loss impact individuals globally, with over 1.5 billion people currently living with some degree of hearing impairment. Despite the medical advancements developed for deaf individuals, there are many limitations, including but not limited to ineffective amplification strategies and hardware inefficiencies. This paper proposes a novel method integrating adaptive thresholding, noise floor detection, and FFT-based analysis to address these limitations. This method is implemented in a wearable haptic feedback system (Vest) designed to translate audio cues into tactile signals in real-time. The potential for practical application in noisy environments and its ability to adjust to fluctuating noise levels set this innovation distinctive. Following successful testing in real-world settings, the haptic motors integrated in this invention reacted appropriately to varying sound levels. A userfriendly graphical interface was also developed to display sound levels and provide user control over sensitivity and intensity settings. These features make this system versatile and effective for everyday use for hard-of-hearing individuals.
Artificial Intelligence-Based Shoreline Monitoring and Hatchling Behavioral Analysis for Sea Turtle Conservation
(Institute of Electrical and Electronics Engineers Inc., 2026-05-21) Palamure, S; Mirahawaththa, G; Krishara, J; Rajendran, K; Vidhanaarachchi, S
Coastal erosion, climate change, and changes in shorelines are serious threats to sea turtle populations. Abnormal behavior in hatchlings raised in hatcheries also affects their survival. Traditional monitoring methods are mainly based on manual observation, which limits continuous and scalable assessment in various conservation areas. Although recent advances in artificial intelligence (AI) show potential in wildlife monitoring and environmental analysis, these issues are often tackled separately. This study introduces an AI-based monitoring framework that combines two related conservation efforts: assessing environmental risks at natural nesting sites and monitoring the behavior of hatchlings in hatcheries. The proposed system merges deep learning for shoreline segmentation with YOLOv8 for hatchling detection. It also includes trajectory based behavioral analysis and spatial risk modeling. This integrated approach allows simultaneous evaluation of environmental changes and hatchling behaviors within a unified analytical framework. Experimental results from video data gathered in coastal nesting areas and hatchery environments show reliable performance. The shoreline segmentation model achieved an Intersection over Union (IoU) of 0.71, with a segmentation mAP@0.5 of 0.71. The hatchling detection model reached a mAP@0.5 of 0.977 and can provide real-time results. By combining environmental and behavioral information, the system supports better risk assessments, including detecting shoreline encroachment and unusual hatchling behavior. In general, this framework offers an efficient and practical solution to automatically monitor sea turtle conservation.
PharmaMap-LLM: Fine-Tuning Large Language Models for Drug-Drug Interaction (DDI) Analysis
(Institute of Electrical and Electronics Engineers Inc., 2025-12-09) Raveendran, K; Sumanathilaka, D; Thiruthanigesan, K; Retnakumar, P; Sivapiragasam, Y
Adverse Drug-Drug Interactions (DDIs) pose a critical challenge under polypharmacy. Traditional methods (rules, classical ML) struggle with scale and novelty in free-text literature. The proposed framework PharmaMap-LLM a DDI extraction framework that instruction-tunes a biomedical LLM via QLoRA for memory-efficient do main adaptation on a single RTX 3090. The system uses an agentic flow: a Predictor (MedLLaMA-3-8B) with two Validators and a clinician-facing Risk agent so that predictions, confidence, and rationales are cross-checked for explainability. Training leverages the DDIExtraction 2013 corpus and an instruction dataset derived from DrugBank pairs (∼20,000), with entity normalization (SciBERT tokenizer). The work reports hardware and training arguments (learning rate, LoRA ranks, batch/sequence sizes) to support reproducibility. On DDIExtraction 2013, PharmaMap-LLM attains F1 (0.87), Precision (0.89), and Recall (0.85), surpassing prior text-mining baselines. Human evaluation by a clinical pharmacologist (200 cases) rated 83% of responses as accurate or better (mean 4.2/5); an LLM-as-judge audit (GPT-4, 2,000 outputs) found 92.3% correctness. Against GPT-4o on 500 shared prompts, validity was similar; the difference w as not significant by McNemar's χ2 (continuity-corrected), p ≈ 0.40. Brief error analysis highlights under-detection of pharmacodynamic patterns and missing dose/time qualifiers, which inform our mitigation strategy (PD-focused exemplars, retrieval of threshold cues, calibrated confidence). Overall, the proposed system achieves an explainable, reproducible, and deployment-friendly DDI tool for clinical decision support.
Optimizing Retail Pharmacy Inventory via Descriptive and Predictive Analysis
(Institute of Electrical and Electronics Engineers Inc., 2025-12-09) Kuruppu, D. S; Galappaththi, K
Effective inventory management is crucial for retail pharmacies to balance supply and demand while minimizing costs. This study proposes an integrated approach combining predictive and descriptive data mining to optimize inventory decisions using sales data from pharmacies in Western Province, Sri Lanka. For predictive analysis, Seasonal Auto-Regressive Moving Average with exogenous variables (SARIMAX) was initially employed but showed limitations in capturing nonlinear trends. Thus, Recurrent Neural Networks (RNN), including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), were implemented, achieving superior accuracy (lowest RMSE: 0.16 for GRU, 0.34 for LSTM). Hybrid models (RNN LSTM + SARIMAX) and ensemble techniques (Bagging, XGBoost) further enhanced performance. For descriptive analysis, K-Medians clustering categorized drugs into 7 clusters based on dosage, price, and weekly sales, revealing key patterns: Cluster 4 contained widely sold essential medications, Cluster 3 comprised high-cost, low-volume specialty drugs, while Cluster 6 represented niche medications with unique demand patterns. The FP-Growth algorithm identified strong purchase associations (support >0.8) between commonly co-prescribed medications. Results demonstrated that RNN-based models outperformed traditional approaches (reducing RMSE by up to 94% versus SARIMAX), while clustering and association rules provided actionable insights for inventory segmentation. This dual analytical framework offers a data-driven solution for inventory management.
Detection and Mitigation of Urban Heat Island Effect using Vision-Language Models
(Institute of Electrical and Electronics Engineers Inc., 2025-12-09) Ayeshmantha S.K.S; Kumara B.D.A.N; Silva G.M.S.S; Madhuwantha G.K.O; Jayasinghearachchi, V; Rajapakse, K; de Silva, R
Urban Heat Islands (UHIs) intensify urban temperatures, contributing to environmental degradation and public health concerns. Mitigating these effects is crucial for creating more sustainable and liveable cities. In this paper, we address the problem of detecting UHI-prone areas and generating context-specific mitigation strategies using street-level imagery. We propose HeatScape, an AI-powered framework that combines semantic segmentation and Vision-Language Models (VLMs) to identify heat-retaining surfaces from user-captured images. Unlike traditional approaches reliant solely on satellite data, HeatScape uses ground-level visual inputs to offer fine-grained urban analysis. VLMs then interpret these segmented scenes to recommend urban planning actions for new and existing building projects. Our experiments show a UHI detection accuracy of 94.25% using logistic regression, with VLM-generated suggestions aligning with established urban sustainability practices. HeatScape demonstrates a scalable, low-cost tool for data-driven environmental assessment and urban design.
The SLIIT Research Document Archive (RDA) is the institutional repository of SLIIT, managed by the SLIIT Library. The primary purpose of SLIIT RDA is to manage, store, and disseminate SLIIT research output with its community and beyond, reaching the wider public. This plays a pivotal role in preserving the academic legacy of the institute.
The collection comprises the research output of SLIIT staff and postgraduate research students, including research publications, conference and symposium papers, books, book chapters, theses, and other scholarly materials. Access to full texts may be restricted depending on the access and licensing terms.