Diagnosis of Common Diseases in Alfalfa (Medicago sativa L.) Plant Using Machine Learning Method and Development of a Mobile Application

Abstract

Alfalfa (Medicago sativa L.), known for its high yield and nutritional value, is a widely cultivated perennial legume subject to various diseases including Alfalfa Mosaic Virus (AMV), Downy Mildew, and Leaf Spot. Timely and accurate identification of these diseases is highly important to maintain crop health, improve productivity, and minimize the use of chemicals. In this study it was aimed to develop a mobile application-based machine learning technique for the detection of major alfalfa diseases. Open-access image dataset of 557 images for four categories—AMV, Downy Mildew, Leaf Spot, and healthy leaves, a deep learning model was used in Google’s Teachable Machine platform. The model then integrated into a mobile application developed with MIT App Inventor 2. The model employs a Convolutional Neural Network (CNN) architecture optimized for mobile deployment via TensorFlow Lite. The application provides a user-friendly interface in Turkish and allows real-time disease classification through mobile phone’s camera. Furthermore, it incorporates cloud-based storage using Google Drive and Google Sheets to log images with metadata including user input, time, and GPS location. The trained model achieved 85% classification accuracy on the test set. The resulting application offers a cost-effective, accessible tool for disease diagnosis in alfalfa cultivation, supporting sustainable agricultural practices. Future studies could expand the application to include a broader range of crops and diseases. The study highlights the potential of integrating artificial intelligence and mobile technology to empower farmers with on-the-spot decision support tools.

Plant phenotyping relevance

アルファルファ葉の画像から病害状態を推定するCNNモデルとモバイルアプリを開発・評価しており、植物病害表現型の取得・分類手法が中心である。

Code and data availability

The paper's image dataset was compiled from the public IPMImages archive and Google Images, and the model was built on Google's Teachable Machine, but no authors' public dataset, code, or trained model deposit is provided. The mobile application is available only by contacting the corresponding author, and no public,论文

No evidence-backed public reproduction asset is currently recorded.