Classification of in-shell peanut kernels nondestructively using VIS/NIR reflectance spectroscopy

One of the grading factors for peanuts is their classification into peanuts with good or bad kernels. Traditional manual methods are labor intensive and subjective. A device by which the classification could be done rapidly and without the need to shell the peanuts would be very useful for the peanut industry. In this work VIS/NIR spectroscopy was used for this purpose. Reflectance spectra were collected for peanut pods (in-shell peanuts) in the wavelength range of 400–2500 nm. A calibration group of about 200 pods were initially scanned to train the classification algorithm. Each individual pod was shelled and the kernels were visually examined and classified as bad if they had any kind of damage, discoloration or immaturity. The remaining pods were marked as good ones. The Principal component analysis model generated from primary spectra with or without pretreatments gave explained variance better than 99%. The maximum normalization model with the ability of characterizing good and bad kernels with an accuracy of 80% and with low SEP and RMSEP values of 0.43, would be useful in the quality characterization of in-shell peanuts.