ASABE Technical Library - Abstract

Abstract. Currently internal bruising of blueberries is evaluated by slicing fruit in half and observing discolored area by human evaluators. This assessment method is time consuming, destructive, and subjective. The goal of this study was to develop a hyperspectral imaging based approach to automatically and non-destructively detect internal bruises of blueberries. A total of 300 blueberry samples from 3 cultivars were divided into four groups: control and bruise treatments dropped from 6, 9, and 12 inches, respectively. The bruise treatments were dropped onto a steel plate using a pendulum, whereas the control group was kept intact. A near-infrared hyperspectral imaging system was used to acquire reflectance images of the blueberries in the range from 950 nm to 1650 nm. Firmness of non-sliced samples and color images of sliced samples were acquired for reference. According to the reference images, mean spectra of healthy and bruised tissues were manually selected to train a support vector machine classifier that was used to obtain classification map. The classification map was fine-tuned by mask images generated using an automated masking method to form final classification images without small noise areas. The final classification images were used to calculate percentage of bruising area of individual blueberries. The classification accuracy of manually selected data achieved 97.10% using 10-fold cross validation, and the trend of the calculated bruising percentages matched with that of the firmness measurements. Therefore, the proposed approach could be an objective and nondestructive way to detect internal bruising of blueberries.