Influence of Illumination on Color based Potato Defect Detection

Abstract

The global vegetable market demands rigorous quality analysis of potato crops to define their reasonable market price. Highly accurate machine learning-based automated potato defect detection systems are necessary for this rising global market. On the other hand, artificial illuminating lamps used in these machine learning systems are decisive in determining their defect detection accuracy. Artificial illuminating lamps used in these machine learning systems should provide a perception of color that is as similar as possible to natural potato color. This paper analysed potato skin color measuring accuracy of seven different illumination lamps (L01 to L07). The analysis used ten potato samples with various defects, including one healthy potato. Experimental results proved that the lowest Δ𝐸∗was obtained when the Compact Fluorescent lamp 45W, 6500 Correlated Color Temperature (CCT) (L1) was used as an illuminating lamp. The proposed experimental study will help to develop machine learning methodologies in future research by generating higher accuracies for potato defect detection using an L01-type illuminating lamp.