Corn Plant Disease Identification Using SURF-based Bag of Visual Words Feature

Feature selection is the important step in image classification due to its influence on accuracy. The objective of this study is to diagnose corn plant diseases using visual features extracted from leaf images with Bag of visual words (BoVW) and the Support Vector Machine (SVM) classification approach. The Speeded up Robust Feature (SURF) approach is implemented to extract and describe the key points of each corn leaf image in the training dataset. The K-Means clustering is utilized to generate k Centroids representing visual words. The arrangement of the BoVW feature based on the histogram of k clusters of visual words provides the input for the SVM classification algorithm. The original contribution of this study is to investigate the impact of number of clusters and proportion of the involved strongest key points toward classification accuracy. The experiment was conducted using the plantvillage public dataset. The experiment results indicate that the best classification accuracy is 85%, with the number of clusters 800 and the proportion of the strongest key points 80%.