Kuhamba, Timothy Kudzanayi. (2020). A deep learning based approach for foot and mouth disease detection. [Unpublished master’s thesis]. University of Zimbabwe.

Abstract

The research gives a new approach for the detection of foot and mouth disease (FMD) using deep learning techniques. The purpose of this study is to provide a timely and accurate early detection of FMD in cattle based on the symptoms. Image data sets of foot and mouth diseased and healthy cattle were collected, preprocessed and different deep learning models were trained to learn features of both healthy and diseased cattle so that these features can be recognised for the classification on images never seen by the deep learning system. There was a difficulty in acquiring diseased cattle images thus a smaller dataset was acquired from the Internet, Veterinary department, European Union Foot and Mouth division (EuFMD) and also Pirbright Institute. Healthy cattle images were taken from the University of Zimbabwe farm with mixed cattle breeds. Different deep learning architectures using transfer learning were assessed and the Densenet 201 outperformed other models with an accuracy of 93.75%, precision 0.98, sensitivity 1.0, specificity 0.9916, AUC 0.99 and ROC of 0.9958. The results also showed the importance of colour information and image focus on the identification of FMD. The study also showed that transfer learning is the best for image recognition when you have a smaller dataset and offers deployable FMD detection system however there is a need for a larger dataset for the detection and identification of each disease symptom. The deep learning system will be used for the development of a mobile application for the detection of FMD. The research under investigation is not intended to replace existing solutions for disease diagnosis, but rather to supplement them.