Fenotipagem de alta eficiência pela análise computacional de imagens no melhoramento genético do tomateiro

• Ano de defesa: 2024
• Autor(a) principal: Sandra Eulalia Santos Faria
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ICA - INSTITUTO DE CIÊNCIAS AGRÁRIAS; Programa de Pós-Graduação em Produção Vegetal; UFMG
• Palavras-chave em Português: Inteligência Computacional; Melhoramento genético de hortaliças; Análise de imagens; Fenotipagem de alto rendimento; Solanum lycopercicum L; Inteligência computacional; Tomate - Melhoramento genético; Biotecnologia
• Link de acesso: http://hdl.handle.net/1843/68232

Tomato (Solanum lycopersicum L.) stands out for its importance in the agricultural and economic scenario, as it is the second most produced vegetable worldwide. Despite biotechnological advances in tomato plant genetic improvement, the integration of innovations, such as computer vision and image analysis, shows promise in accelerating the development of new cultivars. Phenotyping has become more accurate and efficient over the last decade, thanks to this association, enabling large-scale evaluations, a greater number of morphoagronomic descriptors and a reduction in time, human and financial resources. The use of computational analyzes can still facilitate the obtaining of genetic parameters, as well as facilitate evaluations. Therefore, the objective was to establish methodologies for the quantitative and qualitative evaluation of morphoagronomic characters in tomato plants, through computational image analysis. They were carried out two experiments: the first involved commercial lines that were crossed in balanced diallels and conducted until the fruits matured. At this stage, images were acquired at different phases of the crop cycle and seeds were collected; the second experiment, F1 of the first, was constituted of 10 hybrids and five parents, evaluated in the traditional way and through image analysis. For image acquisition, they were obtained videos in the field before harvesting. Furthermore, images of the fruits were acquired in mini-studios using digital cameras and pre-processed in the R software. They were evaluated on the genotype level, characteristics of fruit production and quality such as shape, group, color and defects. It was observed a significant correlation between the characters evaluated in the traditional and computational ways, showing that the use of image analysis, combined with computer vision and deep learning, is an effective tool in tomato plant phenotyping. Furthermore, there was consistency in the estimation of genetic parameters compared to traditional phenotyping. This efficient approach has great potential for tomato plant genetic improvement programs, as it simplifies decision-making and automates phenotyping, reducing time, labor and financial resources.