Transactions of the ASABE Vol. 57(1): 259-272 © 2014 American Society of Agricultural and Biological Engineers ISSN 2151-0032 DOI 10.13031/trans.57.10147 259 CITRUS HUANGLONGBING DETECTION USING NARROW-BAND IMAGING AND POLARIZED ILLUMINATION A. Pourreza, W. S. Lee, E. Raveh, R. Ehsani, E. Etxeberria ABSTRACT. The insect-spread bacterial infection known as citrus greening or Huanglongbing (HLB) is a very destructive citrus disease and has caused massive losses in Florida’s citrus industry. Early, easy, and less expensive HLB detection based on particular symptoms, such as starch accumulation in the citrus leaf, would increase the chance of preventing the disease from being spread and causing more damage. The ability of narrow-band imaging and polarizing filters in detect- ing starch accumulation in symptomatic citrus leaf was evaluated in this study. A custom-made image acquisition system was developed for this purpose in which leaf samples were illuminated with polarized light using narrow-band high-power LEDs at 400 nm and 591 nm, and the reflectance was measured by two monochrome cameras. Two polarizing filters were mounted in perpendicular directions in front of the cameras so that each camera acquired an image with reflected light in only one direction (parallel or perpendicular to the illumination polarization). Four groups of textural features, including gray, local binary pattern, local similarity pattern, and gray-level co-occurrence features, were extracted and ranked us- ing several feature selection methods. Seven classifiers (support vector machine, linear, naive Bayes linear, quadratic, naive Bayes quadratic, Mahalanobis, and k nearest neighbor) were evaluated, and the best classifiers and sets of features were selected based on their accuracy. The leaf samples were collected from the ‘Hamlin’ and ‘Valencia’ varieties of cit- rus. Three classes of samples (magnesium-deficient, HLB-positive zinc-deficient, and HLB-negative zinc-deficient) were considered in the classification process to confirm the starch detection ability of the system. Overall average accuracies of 93.1% and 89.6% in HLB detection were obtained for the ‘Hamlin’ and ‘Valencia’ varieties, respectively, using a step-by- step classification method. The results of this study showed that the starch accumulation in HLB-symptomatic leaves rotat- ed the polarization planar of light at 591 nm, and this property can be effectively used in a fast and inexpensive HLB de- tection system. Keywords. Classification, HLB, Image analysis, Starch concentration, Textural features. itrus is the largest fruit crop in the state of Florida (Putnam, 2012). Based on 2012 statistics, more than 63% of the total U.S. citrus production is grown in Florida. The industry employs nearly 76,000 people directly or indirectly, and it generates almost $1 billion in tax revenues (Brown et al., 2011). One of the harshest diseases affecting citrus production is Huanglong- bing (HLB), which is most likely caused by the insect- vectored α-protobacterium Candidatus Liberibacter asiati- cus (Albrecht and Bowman, 2008). Yield reduction, decline in flavor quality, uneven color development, deformity, and eventual tree death are direct consequences of this disease. An overall 10% loss has been reported due to citrus fruit drop as a result of HLB disease (Choi et al., 2013). In Flor- ida, HLB was first detected in 2005, and it has since spread to all of Florida’s citrus-producing counties. Blotchy mott- led leaves, lopsided fruits with inverted color, yellow shoots, and aborted seeds are some common symptoms of HLB (Gonzalez et al., 2012). To date, no effective treat- ment has been reported for this disease; however, early detection and removal of infected trees can help minimize the spread of the disease to other adjacent trees in the grove. Ground inspection of visual symptoms and polymer- ase chain reaction (qrt-PCR) tests are two common meth- ods of HLB detection. Both methods are costly and time consuming. Fast and easy methods of detection for plant diseases have improved farming productivity and prevented major losses in the agricultural industry. Machine vision is one of the most successful approaches in disease control and de- tection for both trees and herbaceous crops. Fluorescence, hyperspectral, and multispectral imaging, as well as infra- red and nuclear magnetic resonance (NMR) imaging, are different machine vision-based methods that enable farmers to monitor their fields and effectively protect their products from diseases (Sankaran et al., 2010a). Several vision-based methods have been tested for HLB detection. Mishra et al. (2007) reported that the wavelength regions including 530-564 nm (green peak), 710-715 nm (red edge), and two wavelengths (1041 nm and 2014 nm) Submitted for review in February 2013 as manuscript number IET 10147; approved for publication by the Information & Electrical Technologies Division of ASABE in December 2013. The authors are Alireza Pourreza, ASABE Member, Doctoral Student, and Won Suk Lee, ASABE Member, Professor, Department of Agricultural and Biological Engineering, University of Florida, Gainesville, Florida; Eran Raveh, Researcher, Department of Fruit Trees Sciences, ARO Gilat Research Center, Negev, Israel; Reza Ehsani, ASABE Member, Associate Professor, and Edgardo Etxeberria, Professor, Citrus Research and Education Center (CREC), University of Florida, Lake Alfred, Florida. Corresponding author: Won Suk Lee, 1741 Museum Road, Gainesville, FL 32611; phone: 352-392-1864, ext. 227; email: wslee@ufl.edu. C