Estimating the severity of sugarcane aphids infestation on sorghum with machine vision

Abstract

Abstract: Sugarcane aphid (SCA), Melanaphis sacchari, is one of the most prominent insect pests of grain, forage and bio-energy sorghum in the southern US since 2013. The timing and dosage of a pesticide application for SCA depend on a close monitoring of its pressure or severity change in the field. To assist the field scouting, digital images were taken using a smart phone in proximity of infected leaves and corresponding image processing algorithms were developed later to estimate the infestation severity in this study. Image samples were grouped into four classes according to the infestation severity for aphid management considerations: no threat (0-10 SCA/leaf), insecticide use should be considered (11-125 SCA/leaf), insecticide should be used and yield loss likely (126-500 SCA/leaf), and plant death possible (more than 500 SCA/leaf). With 5-fold cross validation, results showed that the best average classification accuracy across the four SCA classes was 85.0% with the modified OVO-SVM algorithm. The SCA quantification accuracies achieved in this study using the SVM algorithm showed the promise of using machine learning algorithms in this case of aphid density estimation on sorghum leaves. The methodology developed in this study can be modified with more sophisticated machine learning algorithms and more data in the future to be incorporated into a handheld or a mobile remote sensing system to assist growers and researchers with automatically quantifying SCA in a fast and objective manner. Keywords: IPM, machine vison, SVM, sugarcane aphid, severity estimation DOI:  10.33440/j.ijpaa.20200302.89  Citation: Deng X L, Thomasson A J, Pugh A N, Chen J X, Rooney L W, Brewer J M, Shi Y Y. Estimating the severity of sugarcane aphids infestation on sorghum with machine vision.  Int J Precis Agric Aviat, 2020; 3(2): 89–96.