Explainable artificial intelligence and deep transfer learning for the diagnosis of maize diseases
Mukisa, Samuel Owekitiibwa
Ntanda, Kimuli Aloysius
Othieno, Benedict Ernest
MetadataShow full item record
In the East African region, agricultural cereal crops such as rice, maize, wheat, and cotton are widely produced. Among these crops, maize is considered to be the most important due to its high nutritional significance enriched with abundant amount of macro-nutrients like starch, fibre, protein and fat along with micro-nutrients like vitamin B complex, ß-carotene and essential minerals. However, diseases have had a significant impact on maize production, and farmers encounter difficulties in managing and diagnosing these diseases, which affects the amount and quality of maize harvests in high yield production. Some of the major maize illnesses include Maize streak virus, Maize leaf blight, Maize lethal necrosis, and Gray leaf spot.In recent years, the use of machine learning techniques such as forecasting, image classification, and object detection, among others, has proven to empower farmers to improve agricultural productivity by performing tasks such as crop production prediction, disease diagnosis, crop and soil monitoring, etc. As a result, this study will employ the use of Explainable AI to carry out maize streak, and leaf blight disease diagnosis. The study will apply supervised Machine Learning (ML) approaches to diagnose maize crop illnesses. This will entail creating and training a custom machine learning model from scratch, followed by Transfer Learning on two pre-trained models, VGG-19 and ResNet50. The accuracy of the designed models will be evaluated using measures such as Precision and Recall, and the top performing model will be chosen. To establish how the model came up with the predictions, the top performing model will be analyzed and explained using Explainable AI techniques. The model will then be deployed on a mobile and web app built using ReactJS and the Django framework.