Classification of defects in potato using grey level co-occurrence matrix and support vector machine

An attempt was made to classify potato into five classes based on features of grey level co-occurrence matrix properties and support vector machine. Potato images were captured using the developed experimental set-up, which had a resolution of 0.22 mm per pixel. The four parameters of grey level co-occurrence matrix viz. variance, correlation, uniformity, and homogeneity were calculated from each image, and the data set was prepared. Sample sizes of crack, rotten, sprout, good, and skin-peel potatoes were 75, 39, 153, 96, and 635, respectively. Cracked, rotten, sprouted, good, and skin-peeled potato images were labelled as 1, 2, 3, 4, and 5, respectively. With addition of labels, the data set size increased to 998-by-five. Feature data set of 998-by-5 was fed into classification learner app in MATLAB. Combined parameters of grey level co-occurrence matrix of variance (contrast), correlation, uniformity (energy), and homogeneity were used in potato classifications. A cubic support vector machine had classification accuracy of 99.5 per cent. Quality parameters like true positive and positive predictive values were studied for cubic support vector machine model. True positive rates of 100, 97, 98, 99, and 100 were observed for crack, rotten, sprouted, good, and skin-peel categories, respectively. Crack and skin-peel categories had same number of potatoes in manual and cubic support vector machine classification. Mis-classified potatoes in cubic support vector machines were less as compared to manual classification by one, two, and two potatoes in rotten, sprouted, and good categories, respectively.