Submitted to: ASABE Annual International Meeting
Publication Type: Proceedings
Publication Acceptance Date: August 25, 2011
Publication Date: August 30, 2011
Citation: Li, X., Lee, W.S., Li, M., Ehsani, R., Mishra, A.R., Yang, C., Mangan, R.L. 2011. Comparison of different detection methods for citrus greening disease based on airborne multispectral and hyperspectral imagery. ASABE Paper No. 1110570. St. Joseph, Mich.: ASABE. Interpretive Summary: Citrus greening, also known as Huanglongbing (HLB), is a serious citrus disease that affects all citrus cultivars and causes rapid deterioration of citrus trees. Early detection is very critical for the management and control of the disease. This study compared several image classification methods for identifying HLB-infected citrus trees from high resolution airborne multispectral and hyperspectral imagery taken from citrus orchards in Florida. Results showed that high resolution multispectral and hyperspectral imagery in conjunction with image classification techniques can identify infected trees with accuracies of 55 to 95% in the validation data sets among the classification methods.
Technical Abstract: Citrus greening or Huanglongbing (HLB) is a devastating disease spread in many citrus groves since first found in 2005 in Florida. Multispectral (MS) and hyperspectral (HS) airborne images of citrus groves in Florida were taken to detect citrus greening infected trees in 2007 and 2010. Ground truthing including ground reflectance measurement and diseased tree confirmation was conducted to build a proper library for HLB infected and healthy canopies. Several classification and spectral mapping methods were investigated to evaluate their applicability to HLB detection. Spectral features derived from both ground reflectance measurement and airborne images were analyzed. Both field, indoor and image spectral analysis showed that HLB infected canopy had higher reflectance in visible range. High positioning error of the ground truth in the 2007 HS image led to detection accuracy of less than 50% in the validation set for every classification method. In the 2010 images, with better ground truth records, more precise library for HLB infected and healthy canopies were collected and higher classification accuracy was then achieved. Spectral angle mapping (SAM) showed the highest detection accuracy of more than 95% in the training sets of both HS and MS images, but its accuracy in the validation set deceased to only 55% in HS image and 62% in MS image. The simpler classification method minimum distance and Mahalanobis distance have somewhat more balanced accuracy rates between the training and validation sets. Support vector machine (SVM) couldn’t work properly in HLB detection, but provided a fast, easy, and adoptable way to build a mask for tree canopy, so that other background could be easily blocked out for classification.