EmotiPot: A Machine-Learning Enhanced Smart Plant System with Visual Diagnosis, Trend-Based Assisted Mobility, and Affective Feedback for Urban Plant Care

Abstract

Many plant owners struggle to maintain healthy plant growth because they rely on manual observation and inconsistent care. Problems such as unnoticed leaf discoloration, wilting, poor light placement, and lack of personalized guidance may lead to plant stress and delayed intervention. To address these concerns, this study developed the EmotiPot system, a smart plant pot integrated with an Android mobile application for real-time plant monitoring and care support. The system combines a MobileNetV2-based Convolutional Neural Network (CNN) for image classification, sensor-based monitoring, content-based filtering for personalized care recommendations, and a sliding-window trend-based method for assisted mobility. The CNN model was trained using 7,830 plant images classified into six categories: blackspots, cancer, greening, healthy, not plant, and wilting. The study followed a developmental research design using the Agile Scrum approach in building and testing the system. Results from the updated model showed a validation accuracy of 93.17%, indicating that the CNN was effective in classifying plant conditions. The system also successfully interpreted soil moisture and light readings, generated suitable care recommendations based on active thresholds, and supported manual and automatic movement toward better light conditions with ultrasonic-based obstacle detection. Overall, the EmotiPot system was able to meet its intended functions and showed potential as a practical and intelligent tool for supporting plant care through real-time monitoring, diagnosis, and guided decision-making.