An Algorithm Oriented to the Classification of Quinoa Grains by Color from Digital Images

The present work proposes an image processing algorithm oriented to identify the coloration of the quinoa grains that make up the different samples obtained from the production of a crop field. The objective is to perform quality control of production based on the statistics of grain coloration, which is currently done manually based on subjective visual perception. This generates results that totally depend on the abilities and the particular criteria of each observer, generating considerable errors in the identification of the colors and tonalities. The problem is further complicated by the nonexistence, at present, of a pattern or standard of coloration of quinoa grains that specifically defines a referential color map. In this sense, through this work, an algorithm is proposed oriented to classify the grains of the acquired samples by their color via digital images and provide corresponding statistics for the quality control of the production. The algorithm uses the color models RGB, HSV and YCbCr, thresholding, segmentation by binary masks, erosion, connectivity, labeling and sequential classification based on 8 colors established by agronomists. The obtained results showed a performance of the proposed algorithm of 91.25% in relation to the average success rate.