AIRBORNE HYPERSPECTRAL IMAGING FOR CITRUS GREENING DISEASE DETECTION

摘要

Huanglongbing (HLB) or citrus greening is the world's most destructive and devastating of all citrus diseases. The disease obstructs the flow of nutrients in citrus trees and HLB infected trees die within 3-5 years. There is no cure for the infected trees, which will have to be removed and destroyed. HLB has now emerged as the major threat to the Florida's $9 billion citrus industry. As of February 2009, citrus trees in 1891 different sections (square mile) in 33 counties were infected in Florida. The disease in Florida is caused by a bacterium, Candidatus Liberibacter asiaticus, that is transmitted by a tiny insect, the Asian citrus psyllid (Diaphorina citri), which thrives on young citrus leaves. Growers urgently need diagnostic tools for early detection, because infected trees may not show symptoms for months or years, during which they are contagious. Current molecular diagnostic tests do not detect the disease soon enough to stop its spread. The objective of this study was to develop a method to detect HLB infected areas in citrus groves using airborne hyperspectral imaging such that it would allow rapid detection of potentially infected areas. This would prevent further spread if followed by development of efficient management plans of these areas. A ground-based inspection, an existing method which is much prevalent and conducted by trained personnel, is subjective, labor intensive and time consuming. On the other hand, airborne hyperspectral imaging would provide much faster results over a wide range of area. An aerial hyperspectral image with a spectral range of 400-1000 nm spanning across 128 spectral bands was acquired from an HLB infected citrus grove in Florida in 2007. The imagery had a 5 nm spectral resolution and 0.7 m spatial resolution. Ground truthing of this area had been carried out and infected tree coordinates were recorded. The images were divided into smaller blocks and sections in order to create training and validation sets. A hyperspectral imaging software (ENVI, ITT VIS) was used for the analysis of these images. Disease infected areas were identified using image-derived spectral library, the mixture tuned match filtering (MTMF), the spectral angle mapping (SAM), spectral feature fitting (SFF), and spectral analyst tool in the hyperspectral imaging software. Result verification and validation required more accurate ground truth information.