Mobile-Based convolutional neural network model for the early identification of banana diseases

Abstract

This study aimed to deploy a deep learning model in a mobile application for the early identification of Fusarium Wilt and Black Sigatoka in bananas. In this paper, a Convolutional Neural Network (CNN) model for the classification of Black Sigatoka banana disease and Fusarium Wilt disease is assessed. A dataset of 27,360 images of diseased and healthy banana leaves and stalks that were collected from the farms using a mobile phone camera served as the training data for this model. An extra class of 407 images that are not of the banana plant downloaded from the internet was used to help the model detect other images not of the banana plant. The CNN model achieved an accuracy of 91.17 % and was deployed in a mobile application for the classification of the diseases. This study shows that deep learning can be implemented and assist in the early identification of banana diseases. The application could detect images of healthy and diseased banana leaves and stalks and images not of the banana plant with a confidence score of more than 90 % in less than five seconds per image and provide research-based mitigation recommendations.

Keywords

• Black sigatoka
• Classification
• Convolutional neural network
• Deep learning
• Fusarium wilt
• Mobile application