Development of an Oil Palm Basal Stem Rot Disease Detection Model Via Machine Vision with Optimized Inception-Based Convolutional Neural Network

Abstract

The oil palm sector plays an important role in the economic model of Malaysia. As plantations expanded, sustainability becomes critical to ensure the fields pass the relevant environmental, social and economic benchmarks. Oil palm trees are susceptible to basal stem rot disease (BSR) caused by the pathogenic fungus Ganoderma boninense. BSR is a fatal and contagious disease which causes a sharp decline in oil palm yields and significantly threatens the production and economic performance of oil palm plantations. Early detection is key to limiting the damage of a BSR infection. Traditional manual detection of BSR is prone to human error and inconsistencies. In response, a number of methods using satellite imagery, spectral imaging and thermal imaging had been introduced with varying degree of success. Alternatively, machine vision (MV) with image processing and feature extraction by deep learning algorithms had been proposed with improved accuracy and feasibility. This paper reviews information on BSR including its stages of infection, symptoms and diagnosis, as well as existing MV models for disease detection in palm-related plants. Additionally, this paper also details the methodology for development of a new MV model for BSR detection based on the Inception convolutional neural network deep learning algorithm.