NIR spectral imaging with discriminant analysis for detecting foreign materials among blueberries

The visualization of foreign materials (leaves and stems) in frozen blueberries was achieved by near infrared (NIR) spectral imaging and discriminant analysis. As a preliminary experiment, NIR spectroscopy of a sample surface was carried out to determine the effective wavelengths for differentiating foreign materials from blueberries in the NIR region. The optimal illumination wavelengths for distinguishing foreign materials were determined to be 1268 and 1317 nm, according to the results of a discriminant analysis of absorbance spectra. Next, absorbance images of areas containing foreign materials and blueberries were acquired by NIR spectral imaging at these two wavelengths. Nine thousand eight hundred and fifty pixels of a blueberry area and 10,107 pixels of a foreign material area were picked randomly from the absorbance images. Discriminant analysis was applied to the absorbance of pixels within the area of interest to determine the discriminant function and threshold value for image binarization. Finally, binary images were obtained by applying the discriminant function and threshold value to each pixel of the absorbance images taken at 1268 and 1317 nm. Foreign materials were clearly distinguished from blueberries as black areas in the binary images. (c) 2010 Elsevier Ltd. All rights reserved.