MEASUREMENT of DISEASE SEVERITY on TOMATO LEAF USING IMAGE PROCESSING

Abstract

Early blight tomato leaf disease is one of the major diseases that affect both quality and quantity of tomato production. Detection and estimation of disease severity so far is carried out traditionaly (using visual observation) that leads to subjectivity. The current research was designed to study the measurement of disease severity of tomato leaf using image processing, the case of Awash Melkasa Research Centre, Oromia Regional State, Ethiopia. The digital image processing techniques were used to detect early blight of tomato leaves by using the area of infected lesion. Ten plot (with plot size of 3*3m 2 ) of tomato plants were selected at Awash Melkasa Research Centre. From each plot, one diseased tomato plant was randomly selected. From each plant six leaves, two leaves were randomly selected from bottom, middle and top parts of every plant. Accordingly, 6 snap shots of TL were taken from each plant for further processing. Totally 60 images representing the samples of tomato leaves were captured. All images were enhanced to improve the contrast between background and foreground, resized to reduce the computational burden and avoid the redundancy. For all images total area of the leaves and total area of disease region of the leaves were extracted. Then after, based on the results of extracted feature, the disease severity of EBT was categorized as I, II, III and IV. Coincidentally, all the upper leaves have fallen in category II (least affected), with regard to middle leaves, nine plants have fallen in category II (least affected) and one plant in category III (moderately affected). For lower leaves seven plants have fallen in category III (moderately affected), two plants were categorized under IV (highly affected) and one plant has fallen in category II (least affected). Early blight tomato leaf disease appears on older leaves (bottom) and goes to the top part of the plant (upper leaf). The accuracy of the algorithm is tested by manual clicking. The experiment proved that the developed algorithm revealed average accuracy of 96.63 %. The results confirm the accuracy (validity) of the system for measurement of the disease severity