Development of a Mobile App for Real-Time Plant Disease Diagnosis Using Image Recognition

Thesis Abstract

The proliferation of plant diseases poses a significant threat to agricultural productivity and food security globally, particularly in regions where access to expert plant pathologists and diagnostic laboratories is limited. The timely identification and management of these diseases are critical for minimizing crop losses; however, traditional diagnostic methods are often time-consuming, costly, and reliant on specialist expertise. This study aims to develop a mobile application capable of real-time plant disease diagnosis through image recognition, leveraging advances in machine learning and mobile computing technologies to bridge the diagnostic gap faced by farmers, agronomists, and extension agents. The primary objectives of this research are to design and implement a user-friendly mobile app utilizing convolutional neural networks (CNNs) for the classification of various plant diseases, evaluate the app’s accuracy and usability in real-world conditions, and assess its potential impact on disease management practices. Specifically, the study seeks to (1) compile and annotate a comprehensive dataset of plant leaf images encompassing multiple crop species and disease types; (2) train and optimize CNN models for high-precision disease classification; (3) develop a mobile application integrating the trained models with an intuitive user interface; and (4) validate the app’s performance through field testing among 150 farmers across diverse agro-ecological zones. A mixed-methods research design was employed, combining quantitative machine learning evaluation techniques with qualitative user experience assessments. The study population comprised smallholder farmers, agricultural extension officers, and plant pathology experts. Sampling involved stratified random selection to ensure representation across different crops and regions, resulting in a sample size of 150 users for field validation. Data collection instruments included a structured questionnaire, semi-structured interview guides, and a set of annotated plant leaf images collected from local farms and existing image repositories. The image datasets, consisting of approximately 10,000 labeled samples, served as the primary data for model training and testing, while user feedback was gathered through surveys and interviews to evaluate app usability and user acceptance. Model performance was analyzed using classification accuracy, precision, recall, and F1-score metrics derived from confusion matrices. The robustness of the CNN models was further assessed through cross-validation and hyperparameter tuning. Additionally, user feedback was subjected to thematic analysis to identify usability strengths and areas for improvement. Results are expected to demonstrate that the optimized CNN models can achieve over 90% accuracy in disease classification under laboratory conditions, with field validation indicating slightly reduced accuracy but acceptable performance levels for practical use. The app’s usability is anticipated to score favorably in terms of ease of use, speed, and perceived usefulness among target users. The study advances knowledge by integrating state-of-the-art image recognition algorithms into a portable mobile platform tailored for resource-constrained environments, contributing to precision agriculture and digital plant health monitoring. It highlights the potential of mobile ICT tools to enhance crop disease management, reduce reliance on expert diagnosis, and promote sustainable farming practices. The key implications include recommending the deployment of such apps in extension services and training farmers in their effective use. Concluding, the research confirms that machine learning-powered mobile applications can significantly improve early detection and management of plant diseases in smallholder farming contexts. Future recommendations include expanding the dataset for broader disease coverage, incorporating multi-language support to improve accessibility, and integrating decision support systems for targeted interventions. Further studies should explore longitudinal impact assessments, scalability, and the integration of additional agronomic advice features to maximize the app’s contribution to sustainable agriculture development.