Journal of Engineering Research and Reports

This study presents the development of a mobile-based intelligent system for the detection and classification of soursop (Annona muricata Linn) leaf diseases using deep learning. Recognizing the limitations of traditional visual inspection methods used by smallholder farmers, the project integrates a convolutional neural network (CNN) model with a mobile application capable of operating fully offline. The CNN, based on MobileNetV2 architecture, was trained using the publicly available SoursopBD dataset containing 3,838 annotated images across six categories: Cutting Caterpillar, Cutting Weevil, Die Back, White Fly, Yellowing, and Healthy. Data augmentation techniques were applied to enhance robustness against real-world imaging conditions. The optimized model, quantized to TensorFlow Lite (TFLite) float16 format, achieved test accuracy of 97.74% with macro averaged precision, recall, and F1-scores of 0.98. When deployed on Android devices, the app-maintained inference times averaging 0.68 seconds per image and remained stable during extended testing, confirming suitability for low to mid-range mobile hardware. Usability evaluation involving 21 participants utilized the UTAUT2 framework, yielding high mean scores for Behavioral Intention (4.79), Performance Expectancy (4.76), and Price Value (4.74). Slightly lower but still positive scores were observed in Trust in Diagnosis (4.63) and Facilitating Conditions (4.55), indicating opportunities for feature refinement and user support. The study concludes that the developed system is technically robust, operationally efficient, and socially acceptable for real-world agricultural deployment. It demonstrates the feasibility of using AI-powered, mobile first tools to enhance plant disease management in resource-constrained environments.