RGB color imaging technique for grading of dates

Abstract

Hardness is one of the major factors in determining the quality of dried fruits. It increases the chewiness and toughness of the fruits. A robust quality assurance system is required for on-line grading of dried fruits as the present manual methods are inconsistent, inaccurate and laborious. The objective of this study was to determine the efficiency of a RGB color imaging technique to classify dates into three classes based on hardness: hard, semi-hard and soft dates. Dates from three common varieties in Oman (Fard, Khalas and Naghal) were used in this study (total 3300 samples). The RGB image of individual date sample was taken by a CCD camera and analyzed using Matlab software. Thirty nine features (13 features in each R, G and B channel) were extracted from each image and analyzed. Three classes (hard, semi-hard and soft) and two classes (hard and soft (“semi-hard and soft” together as “soft”)) classification models were developed using linear discriminant analysis (LDA) with all features and stepwise discriminant analysis (SDA) with selected features (based on level of contribution to classification). In three classes approach, the overall classification accuracy was 69%, 87% and 82% for Fard, Khalas and Naghal varieties, respectively, using LDA. It was 68%, 86% and 81% for Fard, Khalas and Naghal varieties, respectively, using SDA. The classification accuracy was improved in two classes approach. It was 84% (LDA) and 83% (SDA) for Fard, 90% (LDA) and 91% (SDA) for Khalas, and 96% (both in LDA and SDA) for Naghal varieties. Imaging techniques have great potential to develop on-line quality monitoring systems for dates based on hardness. However, further studies are required using other image acquisition systems such as NIR cameras to improve the classification.