Dual-beam spectral measurement improves accuracy of nondestructive identification of internal rot in onion bulbs

Abstract Visible and near-infrared transmittance spectroscopy is useful for detecting internal rot in onion bulbs, but it has limitations for accurate identification of bulbs with mild rot occurring locally in the neck region. We have developed a dual-beam system that measures the transmission spectra in two directions: the neck region (beamline 1: BL1) and across the equator (beamline 2: BL2) when placing an onion with its neck facing upward. Binary classification based on rot scores predicted by partial least squares regression (PLSR) models revealed that the sensitivity of detecting mildly rotten bulbs was superior in a PLSR model developed using the BL1 spectra, whereas the specificity for detecting sound (non-rotten) bulbs and the sensitivity for moderately-to-severely rotten bulbs were better in a PLSR model built using the BL2 spectra. A combined BL1 & BL2 spectra model, constructed using spectra in both directions simultaneously, retained the advantages of both spectra and improved the accuracy of detecting sound and mildly rotten bulbs. Furthermore, we developed a rot-detection threshold optimization procedure that minimized the sum of rates at which decayed bulbs were misidentified as being sound and sound bulbs were misidentified as being decayed, resulting in improved diagnostic performance under real-world conditions.