Near-infrared hyperspectral imaging for detection of bacterial fruit blotch in watermelon seedlings

This study was conducted to detect bacterial fruit blotch (BFB) in watermelon seedlings using near-infrared hyperspectral imaging (HSI) and to compare discrimination accuracies using a partial least squares-discriminant analysis (PLS-DA) model with two extraction methods, band ratio, and principal component analysis (PCA). Three concentrations (1 × 10², 10⁴, and 10⁶ cfu/mL) of Acidovorax avenae subsp. citrulli were inoculated on the first true leaves of watermelon seedlings. Hyperspectral images were captured for 100 samples at 1-5 days after inoculation. The mean spectra for watermelon leaves infected with BFB were higher than those for the healthy leaves in the near-infrared range. Mean spectra for samples increased with increasing days after inoculation. This result was distinctly observed at the higher concentration. Two wavelengths, 1,588 and 1,450 nm, were selected as significant bands for the extraction of diseased area based on statistical t-tests. BFB symptoms were observed even at 1 day after inoculation from hyperspectral images taken out by two extraction methods. A PLS-DA model with band ratio extraction revealed better classification accuracies to distinguish BFB in watermelon leaves. A PLS-DA plot showed distinct groupings of diseased and healthy leaves in the calibration and validation sets. Hyperspectral imaging of watermelon leaves infected with BFB were visualized by applying the PLS-DA model. Our results demonstrate that near-infrared HSI is a promising technique for the detection of BFB in watermelon leaves.