Hyperspectral Reflectance Imaging for Detecting Citrus Canker Based on Dual-Band Ratio Image Classification Method

Li Jiangbo,Rao Xiuqin,Guo Junxian,Ying Yibin
DOI: https://doi.org/10.1117/12.867065
2010-01-01
Abstract:Citrus are one of the major fruit produced in China. Most of this production is exported to Europe for fresh consumption, where consumers increasingly demand best quality. Citrus canker is one of the most devastating diseases that threaten peel of most commercial citrus varieties. The aim of this research was to investigate the potential of using hyperspectral imaging technique for detecting canker lesions on citrus fruit. Navel oranges with cankerous, normal and various common diseased skin conditions including wind scar, thrips scarring, scale insect, dehiscent fruit, phytotoxicity, heterochromatic stripe, and insect damage were studied. The imaging system (400-1000 nm) was established to acquire reflectance images from samples. Region of interest (ROI) spectral feature of various diseased peel areas was analyzed and characteristic wavebands (630, 685, and 720 nm) were extracted. The dual-band reflectance ratio (such as Q720/685) algorithm was performed on the hyperspectral images of navel oranges for differentiating canker from normal fruit skin and other surface diseases. The overall classification success rate was 96.84% regardless of the presence of other confounding diseases. The presented processing approach overcame the presence of stem/navel on navel oranges that typically has been a problematic source for false positives in the detection of defects. Because of the limited sample size, delineation of an optimal detection scheme is beyond the scope of the current study. However, the results showed that two-band ratio (Q685/630) along with the use of a simple threshold value segmentation method for discriminating canker on navel oranges from other peel diseases may be feasible.
What problem does this paper attempt to address?