Disease classification in maize crop using bag of features and multiclass support vector machine

Abstract

Agricultural production is affected by the infection of different pathogenic agents in the various crops. In order to improve productivity for benefitting growing population, early diagnosis and control of diseases using modern technology become important. Maize is one of the important agricultural crops which is affected by various diseases namely Cercospora leaf spot, common rust, leaf blight, etc. The diseases in the crop are identified by recognizing the symptomatic patterns on the leaves and image processing techniques are widely used for classifying such symptoms. In order to accomplish the task, 2000 visible images of maize leaves were obtained from the open access PlantVillage image database. The images are processed to obtain the bag of features and statistical histogram based textural features. The classification of diseases with the obtained features is done using multiclass support vector machine. This study also explored gray level co-occurrence matrix based textural features for the classification of diseases under the various configuration of the multiclass support vector machine. Classification using the bag of features yielded an average best accuracy of 83.7% while using combined statistical features yielded 81.3%. Attributed reason for increase or decrease in accuracy of identification of specific disease type and healthy leaf were also presented.