Rapid analysis of dyed safflowers by color objectification and pattern recognition methods

Abstract

Objective: Rapid discrimination of three classes of safflowers, dyed safflower, adulterated safflower, and pure safflower using computer vision and image processing algorithms. Methods: A low cost computer vision system (CVS) was designed to measure the color of safflowers in the RGB (red, green, blue), L∗a∗b∗, and HSV (hue, saturation, vale) color spaces. The color moments in these three color spaces were extracted from the acquired images as color features of safflower. In addition, five kinds of pigments that are commonly used to dye safflowers were identified by high-performance liquid chromatography as a reference. Pattern recognition methods were investigated for rapid discrimination, including an unsupervised principal component analysis (PCA) algorithm and a supervised partial least squares discriminant analysis (PLS-DA) algorithm. Results: The mean error (e¯) between color values measured with the colorimeter and calculated with the CVS was 2.4%, with a high correlation coefficient (r) of 0.9905. This result indicated that the established CVS was reliable for color estimation of safflowers. The PLS-DA model, which had a total accuracy of 91.89%, outperformed the PCA model in classifying pure, adulterated, and dyed safflowers. Conclusion: The color objectification is a promising tool for rapid identification of dyed and adulterated safflowers.

Keywords

• Safflower
• Coloration
• Adulteration
• Computer vision
• PCA
• PLS-DA