Automatic feature selection using gray-level co-occurrence matrix and detection of mustard plant disease using feed forward neural networks

Abstract

The Mustard crop is one of the important oil seed crops, but the crop suffers from various diseases causing loss in its yield. Diseases can be detected by the naked eyes by experts, but this method is time consuming when applied to large farms. The recent trends in computer vision with the help of various techniques show considerable results in detecting diseases in various crops. As a result, this paper presents a system for feature selection and classification of leaves for detecting diseases at an early stage to prevent the plants from further damage. For image segmentation OTSU technique is applied, GLCM feature selection is used for extracting the features and at last the feed forward ANN classifier is used to classify the infected mustard leaves. The performance of ANN is tested with various other classifiers like RFGBM, J48, Naïve Bayes, Decision Tree, KNN, SVM and RF. Classification parameters, such as Precision, Recall, F-measure, Error, and Accuracy is used as a measure for effectiveness of the feed forward ANN achieving an accuracy of 86.78%.