Abstract
Surface or external qualities of fresh and dried fruits are the important factors in determining the consumer acceptability. An automated and objective method to detect the surface defects on fruits would be highly beneficial in handling and processing facilities. The objective of this study was to determine the efficiency of a computer vision system with RGB color camera to detect the surface cracks on dates. Three grades of dates (no-crack dates, low-crack dates and high-crack dates) were obtained from two commercial dates processing factories in Oman. After the confirmation of grade standards by a dates-quality-expert, the samples were imaged individually using a color camera (105 dates in each grade). Eleven features were extracted from each image and used in classification models. Red, hue and value intensities of three grades of dates were significantly different from each other. In a three classes model, the classification accuracy was 62%, 58% and 78% for high-crack, low-crack and no-crack dates, respectively using linear discriminant analysis (LDA). LDA yielded a classification accuracy of 88% and 75% for the dates with-crack and without-crack, respectively in a two classes model. In pairwise discrimination, the highest classification (96%) was achieved between high-crack and no-crack dates, and the lowest accuracy (59%) was between low-crack and high-crack dates.
