2 nd International Congress of Engineering and Natural Sciences Studies, Ankara, Turkey, 7 - 09 May 2022, pp.201
Abstract Oil and carotenoids are quality components adding value to corn grain. Breeding for these components necessitates analyses of samples from thousands of different genotypes, which requires an extensive amount of labor, time, and money. Although recent developments in image processing technologies allow significant savings, to obtain reliable data, relations between the concentrations of the variables and the images created by them must be determined precisely. Based on the facts that the carotenoids give the maize grain its yelloworange color and that there is a strong correlation between the embryo size and the amount of oil, it seems possible that image processing methods can be used to estimate the amounts of these two variables in the grain. The objective of this study was to determine the carotenoid and oil content in the maize kernel by using image processing techniques. Fifteen maize varieties with different colors and sizes were used, and each was studied with 10 replications. The images of kernels were taken with 3 different cameras, namely multispectral, RGB and infrared. Thereafter, the samples were ground, and the oil extraction was carried out in a soxhlet device. After calculating the amount of oil, absorbance values were taken at 450 nm on the oil + hexane samples using a spectrophotometer to estimate carotenoid amounts. In the image processing, first, an algorithm was designed to work on the images from the maize kernel, which worked in HSV colorspace. In this colorspace, a filter was used to separate the maize kernel image from the background. After that, using the gradient differences with the pixels, the embryo area was separated from the rest of the seed. The color values were calculated from the average hue value, and the area values were calculated based on the pixel count within the embryo. Results indicated that carotenoids and color values had a linear positive correlation, while no correlation could be detected between the oil ratio and embryo size. Carotenoid values could be estimated with an average of 11.96% error.
Keywords: Image Processing, HSV, Multispectral, Maize, Caretenoid