Leaf Disease Recognition Using Ensemble of Laws’ Masks Filters

Abstract

Despite a sharp rise in population, agriculture nevertheless provides food for all people. The control of plant diseases is essential for the preservation of the agroecosystem and the safeguarding of food. To recognise plant diseases, computer-aided methods using artificial intelligence are required. Many techniques previously have been discussed on how to recognise plant disease effectively. One of the methods that are popular to increase the accuracy of the recognition is by using an ensemble approach. Histograms of oriented gradient (HOG) have been used widely as features in detecting disease from leaves. However, HOG is susceptible to rotation and hence makes the recognition accuracy low.  In order to overcome the problem, Laws’ mask filters have been applied towards the images of leaves that have been affected by disease. For each Law’s mask filter, HOG features are extracted from the leaf and classified by classifier support vector machine (SVM). For each Laws’ mask filter, posterior probability gained from SVM is given weight and then added for a particular class. The highest probability obtained from this method is a chosen class for a particular image. The method is applied to 4062 images of grape leaves that contain several types of diseases and healthy leaves. The accuracy of recognition is encouraging, and it serves as a solid indication that the ensemble of Laws' mask filters can aid in enhancing recognition performance.