A Comprehensive Investigation to Cauliflower Diseases Recognition: An Automated Machine Learning Approach

Abstract

Vegetables, a significant part of agriculture, are necessary for the general good health of human beings. The use of information technology can help vegetable farmers to reach high yields which can contribute to global food security and sustainable cultivation. Cauliflower (Brassica oleracea var. botrytis) is a popular vegetable that is easily affected by various diseases causing loss of production and quality. However, machine learning-based disease recognition has yet to be developed for cauliflowers which can help farmers to identify cauliflower diseases and enable them to take timely actions. In this paper, an online machine vision-based expert system for recognizing cauliflower diseases is proposed, where a captured image via a smartphone or handheld gadget is processed and then classified to identify disease to assist the cauliflower farmers. Based on the feature extraction, the system classifies four types of diseases namely ‘bacterial soft’, ‘white rust’, ‘black rot’, and ‘downy mildew’ in cauliflowers. A total of 776 images are utilized to implement this experiment. K-means clustering algorithm has been applied on captured images to segment the disease-affected regions before two-type features extraction namely statistical and co-occurrence feature. Six classification algorithms namely BayesNet, Kstar, Random Forest, LMT (Logistic Model Tree), BPN (Back propagation neural network), and J48 were used for disease classification, and we evaluated their performance using seven performance metrics. We found the Random Forest classifier outperforms all other classifiers for cauliflower disease recognition with accuracy approaching 89.00%.