Garcia Ruiz, Francisco Jose4; Sankaran, Sindhuja5; Maja, Joe Mari5; Lee, Won Suk5; Rasmussen, Jesper6; Ehsani, Reza5
1 Section for Crop Sciences, Department of Plant and Environmental Sciences, Faculty of Science, Københavns Universitet2 Department of Agriculture & Ecology, Crop Science, Department of Agriculture & Ecology, Faculty of Life Sciences, Københavns Universitet3 University of Florida4 Department of Agriculture & Ecology, Crop Science, Department of Agriculture & Ecology, Faculty of Life Sciences, Københavns Universitet5 University of Florida6 Section for Crop Sciences, Department of Plant and Environmental Sciences, Faculty of Science, Københavns Universitet
Huanglongbing (HLB) or citrus greening disease is one of the most important diseases affecting citrus orchards in Florida and other parts of the world. The first critical step for a successful control of HLB is its detection and diagnosis. Spectroscopy has proven to yield reliable results for its early detection, minimizing the time consumed for this process. This study presents a new approach of high-resolution aerial imaging for HLB detection using a low-cost, low-altitude remote sensing multi-rotor unmanned aerial vehicle (UAV). A multi-band imaging sensor was attached to a UAV that is capable of acquiring aerial images at desired resolution by adjusting the flying altitude. Moreover, the results achieved using UAV-based sensors were compared with a similar imaging system (aircraft-based sensors) with lower spatial resolution. Data comprised of six spectral bands (from 530 to 900 nm) and seven vegetation indices derived from the selected bands. Stepwise regression analysis was used to extract relevant features from UAV-based and aircraft-based spectral images. At both spatial resolutions, 710 nm reflectance and NIR-R index values were found to be significantly different between healthy and HLB-infected trees. During classification studies, accuracies in the range of 67–85% and false negatives from 7% to 32% were acquired from UAV-based data; while corresponding values were 61–74% and 28–45% with aircraft-based data. Among the tested classification algorithms, support vector machine (SVM) with kernel resulted in better performance than other methods such as SVM (linear), linear discriminant analysis and quadratic discriminant analysis. Thus, high-resolution aerial sensing has good prospect for the detection of HLB-infected trees.
Computers and Electronics in Agriculture, 2013, Vol 91, p. 106-115