Framework for deep learning diagnosis of plant disorders in horticultural crops: From data collection tools to user-friendly web and mobile apps

Abstract

Food security is a pressing global concern, particularly highlighted by the United Nations Sustainable Development Goal 2 (SDG 2), which focuses on enhancing the productivity and incomes of smallholder farmers. In the Middle East and North Africa (MENA) region, horticultural crops are increasingly threatened by pests and diseases, exacerbated by climate change. Local farmers often lack the necessary expertise to effectively manage these issues, resulting in significant reductions in both yield and quality of their crops. This study seeks to develop an accessible mobile crop diagnosis application. By utilizing machine learning and deep learning technologies, the app is designed to help MENA farmers quickly and accurately identify and treat crop disorders. We used Open Data Kit (ODK) to gather a large dataset of crop images required to train deep learning models. These models, built on open-source deep learning architectures, were designed to classify 21 different leaf disorders, including diseases, pests, and nutritional deficiencies. The system was implemented in both a web app and an Android mobile app. Our deep learning models demonstrated an overall accuracy of 94 % in diagnosing plant disorders. The app, Doctor Nabat, includes a decision support system that offers treatment options in the three primary languages spoken in the MENA region. Doctor Nabat is an effective and scalable tool for enhancing crop management in the MENA region, promoting food security by minimizing crop losses through improved pest and disease diagnosis and treatment strategies.