Faculty of Computing-Scopus

Permanent URI for this collectionhttps://rda.sliit.lk/handle/123456789/4892

Browse

Now showing 1 - 20 of 80

Embargo
A BI Approach for Student Engagement and Retention along with Cognitive Load Analysis for Educator
(979-833153098-3, 2025) Algewatta, M. N; Manathunga, K
This research presents a systematic approach to monitoring student engagement, retention, and cognitive load within higher education by integrating Business Intelligence (BI) tools with cognitive load analysis. The proposed framework utilizes a diverse range of data sources -including attendance, academic performance, mental health indicators, demographic variables, and student feedback to generate real-time insights into student behavior patterns. The BI system identified critical trends, such as irregular attendance, declining academic performance, and the influence of demographic factors, enabling educators to identify at-risk students and intervene proactively. Additionally, cognitive load analysis was employed to evaluate the mental demands of course content, categorizing learning objectives in alignment with Bloom's Taxonomy. This allowed for the identification of content that could potentially overwhelm students, facilitating adjustments in instructional complexity. The integration of BI insights with cognitive load data provided a holistic approach that not only enhanced the monitoring of student engagement but also supported the tailoring of instructional content to optimize learning without inducing cognitive overload. The findings suggest that combining BI tools with cognitive load metrics offers a robust framework for both improving student retention and assisting educators in creating a balanced, engaging, and supportive learning environment. This study contributes a practical model for institutions seeking to leverage data-driven insights to promote student success and address the dynamic challenges of modern higher education.
Embargo
A Dual-Branch CNN and Metadata Analysis Approach for Robust Image Tampering Detection
(Institute of Electrical and Electronics Engineers Inc., 2025) Zakey, A; Bawantha, D; Shehara, D; Hasara, N; Abeywardena, K.Y; Fernando, H
Image tampering has become a widespread issue due to the availability of advanced tools such as Photoshop, GIMP, and AI-powered technologies like Generative Adversarial Networks (GANs). These advancements have made it easier to create deceptive images, undermining their reliability and fueling misinformation. To address this growing problem, we propose a hybrid approach for image forgery detection, combining deep learning with traditional forensic techniques. Our study integrates a dual-branch Convolutional Neural Network (CNN) with handcrafted features derived from Error Level Analysis (ELA), noise residuals from the Spatial Rich Model, and metadata analysis to enhance detection capabilities. Metadata analysis plays a crucial role in identifying inconsistencies in image properties such as timestamps, geotags, and camera details, which often accompany tampered images. The CASIA dataset, a publicly available benchmark for tampered images, was used to train and evaluate the proposed model. After 30 epochs of training, the hybrid method achieved an accuracy of 95%, demonstrating its effectiveness in distinguishing between authentic and tampered images. This research highlights the advantages of combining deep learning models with traditional feature extraction methods and metadata analysis, offering a robust solution for detecting manipulated images. Our findings contribute to advancing image forensics by improving detection accuracy, even in cases involving sophisticated tampering methods driven by AI.
Embargo
A Game Centric E-Learning Application For Preschoolers
(Institute of Electrical and Electronics Engineers Inc., 2025) Kulasekara D.A.M.N.; Nipun P.G.I.; Dombawela H.M.D.L.B.A; Manilka G.S; Manilka G.S; De Silva D.I.
This research explores the potential of advanced technologies such as pose detection (PD), augmented reality (AR), object detection (OD), and voice recognition (VR) in creating a game-centric e-learning application for preschoolers. The proposed application, Kidstac, integrates cognitive and physical development through interactive activities with real world interaction, addressing gaps in traditional e-learning methods that often neglect physical engagement. The app features real-time feedback mechanisms and structured modules like virtual zoo explorations, exercise games, treasure hunts, and pronunciation activities. Testing results indicate significant improvements in motor skills, knowledge retention, problem-solving abilities, and language proficiency. These findings demonstrate the effectiveness of blending physical and digital learning experiences to enhance early childhood education. The study establishes a foundation for scalable, activity-based learning tools, emphasizing the holistic development of young learners.
Embargo
A Reinforcement Learning Approach with Human in the Loop to Explainable Insurance Risk Scoring and Intelligent Policy Portfolio Optimization
(Institute of Electrical and Electronics Engineers, 2026-05-29) Gamage, C; Kasthuriarachchi, T; Denuwan, C; Mallawaarachchi, P; Abeywardhana, L; Nawarathne, M
Assessing individual risk accurately and optimizing insurance portfolios in real time remain major challenges due to static actuarial tables, opaque models, and fragmented analytical pipelines. This paper proposes a two-part Explainable AI (XAI) framework addressing both issues. The first component, Artificial Intelligence-driven risk scoring with human-in-the-loop (HIL) weight adjustment, uses a Proximal Policy Optimization (PPO) agent to suggest feature-based changes to an insurer's risk-equation weights. Shapley Additive Explanations(SHAP) attributions and Generative AI reasoning make these changes interpretable, allowing human reviewers to approve modifications that are immediately applied to future customers, creating a self-improving loop. The second component, AI-driven policy optimization, leverages a PPO supported by an XGBoost expense regressor, SHAP/LIME explainability, PPO agent and a Retrieval-Augmented Generation (RAG) layer for rider assignment. Both components share a data backbone of 100,000 anonymized insurance records stored in MongoDB, enabling incremental updates without reprocessing. Experiments show the XGBoost regressor achieves Root Mean Square Error (RMSE) 0.4406 and Mean Absolute Error (MAE) 0.3600, the HIL guided agent increases average episodic reward by 10.3%, and the RAG layer reaches 91.7% rider-assignment accuracy. The framework significantly enhances predictive accuracy, interpretability, regulatory traceability, and portfolio adaptability compared to traditional actuarial and black-box approaches.
Embargo
A Secure Protocol for Computer-Based Assessments in Disrupted Environments
(Institute of Electrical and Electronics Engineers Inc., 2025) Navin, D; De Zoysa, K; Karunaratna D.D; Harshanath, B
Examinations are fundamental to education, yet conducting secure computer-based exams in disrupted environments presents significant challenges. This research introduces a Secure by Design protocol leveraging Delay Tolerant Networks (DTN) to overcome connectivity gaps in remote and resource-constrained areas. The proposed solution integrates physical, administrative, and technical controls to ensure the confidentiality, integrity, and availability of examination data. Through an iterative action research approach, the system evolved from a centralized Moodle server to standalone local servers, enabling offline functionality and enhanced resilience. Tested across over 180,000 candidates in Sri Lanka's largest computer-based examination, the framework effectively addressed power outages, internet disruptions, and logistical constraints. The findings demonstrate the protocol's effectiveness in promoting equitable and reliable access to education, ensuring examination continuity despite adverse conditions.
Embargo
'AAYU', Paralyze Ease Home Suite and Mobility Companion
(IEEE Computer Society, 2025) Tharushi N.K.; Ranaweera D.G.K.T.T.; Munasinghe A.S.; Wijesekara P.N; Gamage N.D.U; Pandithage D
Ensuring the safety and well-being of paralyzed individuals remains a critical challenge, particularly in resource-limited settings. Limited access to assistive technology and real-time monitoring increases health risks and dependency. This paper presents AAYU (Assistive Automation for Your Upliftment) Paralyze Ease Home Suite and Mobility Companion, an intelligent system integrating home automation and wearable technology to enhance patient safety, communication, and autonomy. AAYU addresses four key challenges: (1) optimizing home environments through automated adjustments based on vital signs, (2) enabling nonverbal communication via a voice-to-text smart device, (3) detecting falls with a real-time positioning belt, and (4) preventing deep vein thrombosis (DVT) using a sensor-equipped monitoring belt. An initial evaluation demonstrates AAYU's potential to improve the quality of life for paralyzed individuals through proactive and adaptive support.
Embargo
Adaptive Robotic Voice Modulation for ASD Kids: Tailored Voice Pitch, Tone, and Speed
(Institute of Electrical and Electronics Engineers Inc., 2025) Panduwawala, P; Pulasinghe, K; Rajapaksha, S
Children with Autism Spectrum Disorder (ASD) often experience sensory sensitivities, particularly auditory hypersensitivity, which can make interactions and communication challenging. This study explores the customization of the NAO robot's voice pitch, tone, and speech speed using the Kaldi Speech Recognition Toolkit to align with the preferences of children with ASD. Eight distinct voice profiles were created, offering a range of variations in pitch, tone, and speech speed. Parents or caretakers were asked to select the voice profile they felt would be most suitable for their child. Based on this feedback, we created a spectrum of voices tailored to each child's needs. Results indicate that medium-pitch and moderate-speed combinations are most effective in enhancing engagement, with Voice 2 emerging as the preferred profile. The findings underscore the potential of adaptive voice modulation in improving robotic interactions for ASD therapy and highlight opportunities for further research in real-time adaptability and long-term impact assessment.
Embargo
Adaptive Voice Communication in Emotion-Aware Digital Companions
(Institute of Electrical and Electronics Engineers Inc., 2025) Rathnayake, P; Rathnaweera, C; Jithma, U; Aththanayake, I; Rathnayake, S; Gunaratne, M
This paper presents an adaptive voice communication system for emotion-aware digital companions that dynamically responds to users' affective states through expressive speech and synchronized 3D avatar animation. The system integrates real-time voice input, emotion recognition, and context-aware dialogue generation using GPT-3.5, followed by emotional text-to-speech synthesis via neural TTS. Lip-sync data is generated using phoneme alignment and rendered in sync with the avatar's facial expressions and gestures. To enhance user trust and engagement, the avatar visually mirrors the emotional tone of the speech. A cultural adaptation layer is introduced to align voice output and speech style with Sri Lankan communication norms, including tone, pacing, and formality. Implemented using a Node.js backend and React + Three.js frontend, the system demonstrates strong potential for emotionally intelligent, culturally adaptive AI interactions. This work contributes a modular pipeline for building empathetic voice agents capable of enhancing realism and trust in human-AI communication.
Embargo
Advancing Speech Therapy for Sinhala-Speaking Children with Autism Spectrum Disorder Through an Intelligent Dialog System
(Institute of Electrical and Electronics Engineers Inc., 2025) Jayawardena, A; Pulasinghe, K; Rajapakshe, S
This paper presents a dialog system integrated with a NAO socially assistive robot, designed to support Sinhala-speaking children with Autism Spectrum Disorder (ASD). The system leverages a pipeline-based architecture implemented using the RASA framework, consisting of Natural Language Understanding (NLU), Dialog Management (DMU), and Natural Language Generation (NLG) units. The NLU unit processes user input by identifying intents, entities, and dialogue acts, incorporating custom tools like the SpokenSinhalaVerbTokenizer for handling spoken Sinhala. The DMU includes a Dialog State Tracker (DST) to maintain conversation context and a Dialog Policy Generator, which employs rule-based, TED, and UnexpecTED policies to adapt conversation flows dynamically. The NLG unit generates natural responses to foster interactive and goal-oriented conversations. Integrated with the NAO robot, the system engages children through meaningful dialogues, such as discussing toy preferences, aiming to enhance social interaction and communication skills. This work highlights the potential of conversational AI and robotics in therapeutic interventions for ASD in low-resource languages.
Embargo
AgriSense LK: Weekly Automated Machine Learning for Sri Lankan Produce Prices with Business Continuity Plan, Market Opportunity Ranking, Cultivation Targeting, and Yield Quality Valuation
(Institute of Electrical and Electronics Engineers Inc., 2026-05-22) Matharaarachchi, Charaka J.; Samarasinghe, Ravindu T; Vidyasarani G.G.T.; Fasnas, M; Siriwardana, D; Wijesooriya, A
In Sri Lanka, agricultural decision-making remains largely traditional: decisions are often based on historical practices, informal consultation, and heuristic judgment. The primary barrier is that market price data is difficult to interpret without analytical expertise, resulting in unpredictable price volatility and suboptimal farmer income. AgriSense LK is a machine learning platform that converts historical price records into actionable recommendations for farmers and traders. The system comprises four components: business strategy classification, market opportunity ranking, cultivation targeting, and smartphone-based produce quality grading. The platform was trained on 123,985 real price records sourced from the Central Bank of Sri Lanka (CBSL), spanning 2017 to 2025. Key results include a MAPE of 0.7% and MAE of Rs. 1.86 on weekly price forecasting (a 98.1% improvement over the naive baseline), a ROC-AUC of 0.9056 on cultivation targeting, and 91.49% crop classification accuracy with 89.84% quality grade accuracy in the computer vision component. Direct price regression over a seven-day horizon proved unreliable; a binary profitability classifier was adopted instead and substantially outperformed the regression approach. While results are promising, further validation under real-world deployment conditions is required.
Embargo
AI Interviews with Facial Emotion Recognition for Real-Time Feedback and Career Recommendations
(Institute of Electrical and Electronics Engineers Inc., 2025) Herath R.P.N.M; Arachchi D.S.U.; Gunaratne M.H.B.P.T.; Hansana K.T.; Wijayasekara, S.K; Jayasinghe, D
The hiring process is complex, requiring evaluation of candidates across multiple dimensions, including technical proficiency, behavioral traits, and credibility. Traditional interviews often suffer from biases and inefficiencies. This research presents an AI-driven Interview System integrating Machine Learning (ML), Natural Language Processing (NLP), and Computer Vision to automate and enhance recruitment. The system generates contextual interview questions, evaluates candidate responses using LLM-based scoring models, and provides real-time feedback for engagement. It includes speech-to-text transcription and offensive word detection to ensure professionalism. The behavioral analysis module leverages facial emotion recognition and computer vision to assess non-verbal cues such as confidence and attentiveness. Additionally, Curriculum Vitae (CV) parsing and LinkedIn data extraction use NLP-based entity recognition to extract educational background, work experience, and key skills, enabling personalized interviews. The technical assessment module administers real-time coding challenges, evaluating solutions for correctness, efficiency, and best practices while providing AI-generated feedback. By automating these key hiring aspects, this system enhances objectivity, efficiency, and decision-making, ensuring a data-driven, unbiased, and scalable selection process while improving the candidate's experience and employer insights
Embargo
AI Powered Integrated Code Repository Analyzer for Efficient Developer Workflow
(Institute of Electrical and Electronics Engineers Inc., 2025) Akalanka, I; Silva, S.D; Ganeshalingam, M; Abeykoon, A; Wijendra, D; Krishara, J
Transitioning between new and legacy codebases in diverse project environments poses significant challenges for developers, especially with traditional Knowledge Transfer (KT) methods, which are often resource intensive and prone to obsolescence. These limitations hinder the Software Development Life Cycle (SDLC), particularly in fast-paced industrial settings. This research introduces an AI-driven automation solution that leverages large language models (LLMs) and advanced artificial intelligence technologies to address critical gaps in technical knowledge transfer, with a focus on modern software frameworks. The proposed system reduces development costs, improves team performance, and accelerates adaptation to complex codebases. Key features include a documentation generation tool that cuts manual effort by up to 90%, with an average generation time of 6.8 minutes. Additionally, a virtual knowledge transfer assistant enhances onboarding efficiency, potentially reducing senior developer involvement by 50-60%. The system also includes an automated diagram generator that achieves 97% validation accuracy and a code smell detection tool with 71% accuracy, resulting in better code quality assessments. These findings demonstrate the effectiveness of AI-driven automation in improving developer productivity, streamlining onboarding processes, and optimizing software development workflow
Embargo
Ai-Based Urine Microscopy Image Analysis for Predicting Urinary Tract and Renal Diseases
(Institute of Electrical and Electronics Engineers, 2026-05-29) Senanayake, K; Panagoda, P; Dharmapriya, S; Chathurya, R; Wijendra, D; De Silva, H; Jayawardana, D
The manual microscopic examination of urine is a crucial step in diagnosis of urinary tract and renal diseases. However, the process is time-consuming and operator dependent. Most of the existing automated urinalysis techniques only consider the microscopic examination of individual components or the black-box-based prediction models. There is a lack of a comprehensive framework that incorporates the microscopic examination of the microscopic components with the clinical diagnostic logic. In this regard, the present work proposes an artificial intelligence-based urinalysis system for the microscopic examination of the components in the urine sample to generate diagnostic outcomes. In the proposed system, the microscopic components like white blood cells, red blood cells, bacteria, yeast, crystals, and casts are detected and analyzed to generate the diagnostic outcomes for the diagnosis of urinary tract infection, kidney stone risk, hematuria causes, and casts-related renal diseases. In the proposed system, efficient lightweight models ensure precision and effectiveness in identifying various biological entities. White blood cells are detected with a mAP@0.5 score of 0.96, yeast with over 0.94, and crystals with more than 0.91, yielding a classification accuracy of 99.41% for crystals. The system detects microscopic elements like casts with a mAP@0.5 score of 0.80. The system also incorporates auxiliary clinical data to enhance diagnostic results for various diseases.
Embargo
AI-Driven Decision Support System for Sustainable Agarwood Cultivation and Export Readiness
(Institute of Electrical and Electronics Engineers Inc., 2026-05-22) Malmali W.S.V.M.O; Jayawardena L.P.G.K.; Rathnamalala R.M.B.I.T.; Kandage T.P.; Weerasinghe, Lokesha
Agarwood cultivation and export involve multiple critical decision points that directly affect resin quality, economic value, and market acceptance. In current practice, decisions related to resin induction timing, disease identification, and export readiness are largely based on manual inspection and subjective judgment, leading to inconsistent assessments and avoidable losses. This paper presents an AI-based decision support system to support sustainable agarwood cultivation and export by integrating three analytical components: resin induction stage classification, export readiness and quality assessment, and leaf disease detection with remedy recommendation. A multimodal deep learning approach combining bark images and numerical tree parameters is used for resin induction stage classification, achieving a test accuracy of 93% using an EfficientNetB0 with a Multi-Layer Perceptron (MLP). Agarwood resin and chip quality grading is performed using an EfficientNetB0-based Convolutional Neural Network (CNN), while export readiness is evaluated using a Random Forest-based numerical model. Leaf disease detection is implemented using a CNN-based classifier, achieving an overall accuracy of 80% across four common agarwood leaf disease classes. Explainability mechanisms, including Gradient-weighted Class Activation Mapping (Grad-CAM) and reason based readiness analysis, are incorporated to enhance transparency and user trust. Experimental results indicate that the proposed system reduces subjectivity and supports data-driven decision-making across key stages of the agarwood value chain.
Embargo
AI-Driven Fault-Tolerant ETL Pipelines for Enhanced Data Integration and Quality
(Institute of Electrical and Electronics Engineers Inc., 2025) Wickramaarachchi, C.K; Perera, S.K; Thelijjagoda, S
The reliability and fault tolerance of ETL (Extract, Transform, Load) pipelines are essential for maintaining data integrity in corporate environments. Traditional ETL systems often depend on manual interventions to resolve data inconsistencies, leading to errors, inefficiencies, and increased operational costs. This study introduces an AI-driven framework designed to improve the fault tolerance of ETL processes by automating data cleaning, standardization, and integration tasks. Using machine learning models, the framework reduces the need for human intervention, enhances data quality, and supports scalability across various data formats. Using real-world data sets, the proposed solution demonstrates its ability to improve operational efficiency and reduce errors within corporate data pipelines. This research addresses a crucial gap in ETL automation, offering a scalable and proactive approach to robust data integration in large-scale corporate settings. The findings highlight the ability of the framework to improve fault tolerance, improve data quality, and offer organizations a competitive advantage in managing complex data ecosystems.
Embargo
AI-Driven Integrated Caregiving and Health Monitoring Framework for Elderly Well-Being
(Institute of Electrical and Electronics Engineers, 2026-07-22) Nugaliyadde, S; Nikeshi, N; Marasinghe, M; Rajapaksha, C; Rajapaksha, S; Thelijjagoda, S
The rapid growth of the aging population has brought about some serious challenges, particularly in managing illnesses, feelings of loneliness, cognitive decline, and mental health issues. Traditional caregiving methods often depend on occasional assessments and hands-on supervision, which can fall short in providing the ongoing and adaptable support that’s really needed. This paper introduces an innovative caregiving and monitoring framework powered by AI, aimed at offering integrated, real-time, and comprehensive assistance for older adults. The system harnesses the power of Artificial Intelligence (AI), Machine Learning (ML), Natural Language Processing (NLP), and health data analytics to combine physical health monitoring, nutrition planning, smart routine coaching, and therapist-led mental health support all in one platform. With features like voice-based conversations and journaling, it makes emotional expression and behavioral analysis more accessible, helping to gain a deeper insight into users’ mental well-being. Predictive analytics and anomaly detection are used to spot early signs of health risks and shifts in behavior, allowing for timely interventions. Plus, remote access means caregivers and healthcare professionals can keep an eye on users and offer informed advice. By shifting caregiving from a reactive approach to a proactive and preventive one, this system not only improves quality of life but also encourages independent living and eases the burden on caregivers.
Embargo
AI-Driven Vehicle Valuation and Market Trend Analysis for Sri Lanka's Automotive Sector
(Institute of Electrical and Electronics Engineers Inc., 2025) De Silva K.P.N.T.; Shehan H.A.; Jayawardhane A.S; Premarathne A.P.S.; Krishara, J; Wijendra, D.R
The automotive sector in Sri Lanka faces challenges in vehicle valuation accuracy and market trend analysis due to fluctuating prices, varying vehicle conditions, and environmental concerns. This paper presents an AI-driven vehicle valuation system integrating machine learning models for automated vehicle identification, damage detection, market trend analysis, and environmental sustainability assessments. Using deep learning techniques such as Convolutional Neural Networks (CNNs) and time-series models like Long Short-Term Memory (LSTM), the system delivers accurate valuation and market trend insights. Experimental results demonstrate 9 2% accuracy in damage classification and a mean absolute error (MAE) of 5.3% in repair cost estimation, supporting informed decision-making. This research bridges gaps in valuation transparency and sustainability in emerging automotive markets.
Embargo
An Adaptive E-Learning Platform for Individuals with Down Syndrome
(Institute of Electrical and Electronics Engineers Inc., 2025) Sandaruwan U.V.S.; Dias A.H.J.S.S; Shamindi H.M.H; Priyawansha N.G.D.; Chandrasiri L.H.S.S; Attanayaka B.
Children with Down Syndrome (DS) encounter varying degrees of learning disabilities within the traditional education framework, requiring personalized interventions. This paper presents Blooming Minds, an adaptive, Machine Learning (ML) driven e-learning platform designed to support the development of cognitive, linguistic, and motor skills in children with DS. Built on the VARK (Visual, Auditory, Reading/Writing, Kinematic) theory, the platform provides personalized activities using real-time feedback mechanisms. The system includes nine interactive modules that cover the above VARK theory. It uses ML algorithms, including Support Vector Machine (SVM) and Random Forest (RF) for screening, Convolutional Neural Networks (CNN) for handwriting and speech analysis, Long Short-Term Memory (LSTM) for sequence prediction, and Reinforcement Learning (RL) for adaptive difficulties. Handwritten letters and voice samples from children with DS, both domestic and international, were specifically considered as inputs for this research. Progress tracking dashboards provide visual insights for educators, parents, and caregivers, improving support and adaptability. The system achieved 91.26% accuracy in letter recognition and 88% in speech classification. This e-learning platform has been recognized as an effective solution in Sri Lanka, allowing for further correlations and investigations to assess the knowledge capacity and ability to express that knowledge in children with DS
Embargo
An Ethical and Emotionally Intelligent Social Media Plugin Using Responsible & Explainable AI
(Institute of Electrical and Electronics Engineers Inc., 2026-08-04) Wickremasinghe N.S.; Ullandupitiya U.P.L.I.; Jayawardena D.S; Jayanetti J.K.D.S.D; Rathnayake S.; Nawarathne M.
Social media platforms encounter ongoing difficulties with fragmented moderation and interaction operations that do not collaborate effectively to deal with destructive content, inauthentic behavior, emotion-insensitive interaction, and opaque recommendations. This paper features an ethical and emotionally intelligent social media plugin using responsible and explainable AI implemented as a unified real-time service for social media applications, proven through implementation on the Open-Source Social Network (OSSN). The plugin incorporates four elements, namely multimodal cyberbullying detection, behavior-based social bot detection, explainable friend recommendation, and an emotion-aware reaction system. The cyberbullying module is a combination of transformer-based text analysis, image processing, OCR, and keyword fusion to moderate content in an explainable manner, and the fake account module uses temporal behavioral features to detect bot accounts regardless of content. The recommendation system offers interpretable recommendations with context, and the emotion-aware module provides empathetic interaction via emotion recognition and filtering. Experimental results indicate high performance across task-appropriate metrics such as accuracy for classification tasks and MCC/ROC-AUC for bot detection due to class imbalance, yielding 88.60% on text-based cyberbullying, 68.81% on image moderation, MCC 0.9704 and ROC-AUC 0.9981 on bot detection, and 73.10% F1 on sarcasm detection. The outcomes of deployments demonstrate the potential of a single, transparent, and responsible AI system to have safer and more meaningful social media interactions
Embargo
An Explainable Deep Learning Framework for Coconut Disease Detection Using MobileNetV2, Super-Resolution, and Grad-CAM++
(Institute of Electrical and Electronics Engineers Inc., 2025) Balasooriya R.C.; Adithya E.L.A.Y; Gunarathne M.M.S.U; Silva T.C.D; Lokuliyana, S; Wijesiri, P
Coconut production is a significant industry in Sri Lanka's economy and food security. However, it is constantly under threat from diseases such as Grey Leaf Spot and pests such as Coconut Mites (Aceria guerreronis). Detection must be early, but it is difficult, especially in field conditions where image quality is low and symptoms are not visually distinguishable. This paper proposes a two-stage deep learning solution to enhance and automate disease and pest recognition with a lightweight and mobile system. The system combines Real-ESRGAN based image super-resolution to restore visual detail in poor-quality mobile images and MobileNetV2-based classification, a lightweight convolutional neural network. The model recognizes grey leaf spot with over 97% accuracy and greatly enhanced mite recognition performance when combined with super-resolution preprocessing. In the interest of transparency and trust for users, the Grad-CAM++ and LIME interpretation techniques are utilized, and visual explanations of the predictions are presented. A mobile application was created with React Native and integrated with a Flask-based backend to enable real-time image enhancement and classification to facilitate practical deployment. Smartphone-captured field-level photos were preprocessed and categorized into healthy, diseased, and non-coconut samples. Farmers can use the proposed system in real time because it maintains good accuracy while being computationally efficient. This framework provides a scalable method for intelligent and sustainable agriculture.