Saleh M. Ashaghathra and P. Weckler, Oklahoma State University. Multi-method recognition of pecan weevils. Applied Imagery Pattern Recognition (AIPR) 2007 workshops
Pecan Weevil is one of the most destructive pests of pecans and are considered as the most serious �late-season� pest because they attack the nut. Nut losses due to pecan weevils can cause significant financial loss and can cause total crop failure. The goal of this study is to identify pecan weevil among other insects that are naturally present in the pecan habitat by implementing several image processing techniques. This is the first step toward building a wireless imaging system that would be commercially available for farmers. Insect recognition based on template matching was used in this study. Over 205 pecan weevil insects were used as a training set. The testing set consisted of 75 insects which normally exist in pecan habitat as well as 30 pecan weevils. Correlation-based template matching and geometrical image descriptors were the two recognition procedures used. Four types of geometrical descriptors, namely, Fourier descriptors, Zernike moment invariants, string matching and regional descriptors were used. The algorithm consisted of applying the above five methods. Moment invariants gave the lowest type I error; however, the type II error for this method is very large. NCC and regional descriptors methods gave the lowest type I and type II error. Further, string matching and Fourier descriptors produced the highest type I error. Results indicated that a positive match from three of the five independent tests would yield reliable results. Therefore, the algorithm included all five methods as no single method would achieve the desired success rate for identifying pecan weevils.