Fenotipagem por imagens: uma alternativa às avaliações visuais no melhoramento genético de batata

Abstract

Flesh color and tuber shape are important traits for market segmentation in potato. In breeding programs, these traits are usually evaluated visually, which may compromise experimental precision. Therefore, this study aimed to develop and validate the use of image-based evaluation for estimating genetic parameters related to tuber flesh color and shape. Two populations from the Potato Breeding Program of the Federal University of Lavras (UFLA) were evaluated in 2023 and 2024 at the UFLA experimental farm in Lavras, MG, using a partially replicated design (P-REP). After harvest, tubers were evaluated for flesh color and shape. Images from each plot were analyzed to obtain the mean b* channel value from the LAB color space. For validation, b* channel data for each treatment were correlated with the average scores of three evaluators (NCP). To assess shape, all tubers from each plot were photographed, and the major and minor axes of each tuber were obtained through Python processing to calculate the diameter ratio (RD). Validation was performed by correlating the mean ratio for each treatment with the average scores of three evaluators for shape (NFT). Heritabilities, mean standard deviations of contrasts between genotypes divided by the mean of the clones (DPMC, %), coefficients of variation (CV, %), and accuracies were also calculated. Correlations between visual scores and image data were high, above 0.92 for flesh color and 0.78 for shape, validating the use of images for evaluating these traits. Heritability estimates obtained from images were higher than those derived from visual scores, ranging from 0.64 to 0.90 for RD and from 0.75 to 0.86 for the b* channel, demonstrating the greater ability of images to detect genotypic variation. Image-based evaluation also showed higher experimental precision, with DPMCs and CVs up to eleven times lower for flesh color and up to eight times lower for shape. The contribution of genotype × year interaction in joint analyses was minimal, reinforcing the stability of clone performance over the evaluated years. It is concluded that image-based evaluation provided greater experimental precision than visual evaluation, offering more reliable estimates of genetic parameters for flesh color and shape.