Near infrared spectroscopy-based method for separation of sound and insect-damaged seeds of Albizia schimperiana, a multipurpose legume

The potential of near infrared spectroscopy as a rapid method to separate sound and insect-damaged seeds of Albizia schimperiana Oliv is demonstrated. Near infrared reflectance spectra were collected on single seeds of each fraction soaked in water for one, three, six, nine and twelve hours. To separate sound and insect-damaged seeds, multivariate classification models were developed with partial least squares (PLS) regression using the digitized spectra as a regressor and a y-vector of artificial values (1 for sound and -1 for damaged seeds) as regressand. The result showed a 100% classification rate of sound and damaged seeds in test sets after 6, 9 and 12 hours of imbibition based on full spectrum models. A similar classification result was achieved based on selected absorption bands. PLS weights indicated that the origin of spectral differences between sound and damaged seeds is attributed to difference in relative water content. The result predicate the prospect of developing rapid filter-based sorting equipment that can easily be automated.