Feasibility study of utilising electronic nose to detect BSR disease in oil palm plantation

Abstract

The agricultural industry has been, for a long time, dependent upon human expertise to detect plant disease. However, human experts may take years of training and can be inconsistent, as well as prone to fatigue. Presented in this thesis is the work conducted on utilising electronic nose incorporating artificial intelligence to detect plant malaise, specifically, basal stem rot (BSR) disease that is caused by Ganoderma boninense, a type of fungi affecting oil palm plantations in South East Asia. A commercial electronic nose, Cyranose 320, was used as the front-end sensors with artificial neural networks trained using Levenberg-Marquardt algorithm employed for decision making. For the first stage, a study on Cyranose 320 embedded pattern recognitions and artificial neural networks (ANNs) was conducted using a few types of essences. This stage confirmed that the ANNs is better than the embedded pattern recognitions in terms of accuracy and hence should be used for the next experiments. The second stage involved the Ganoderma boninense fruiting bodies detection in laboratory and oil palm plantation. This stage proved that the fungi odour can be detected after being tested using a few types of odour parameter. The next stage is to discriminate the healthy and non-healthy oil palm trunk in the plantation. The conducted work indicates that the combination of the electronic nose and ANNs has the ability to discriminate the infected trunk. The findings of the work were also used to develop an in-house low cost electronic nose to support further fundamental study and implementations. As a conclusion, this work confirms that it is feasible to utilise the electronic nose and ANNs to detect and discriminate the BSR disease both in the laboratory and in the plantation.