Ferramenta computacional para apoio ao gerenciamento e à classificação de sementes de soja submetidas ao teste de tetrazólio

• Autor(a) principal: Rocha, Davi Marcondes
• Data de Publicação: 2016
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do UNIOESTE
• Texto Completo: http://tede.unioeste.br/handle/tede/3086
• topic: Qualidade de sementes; Vigor de sementes; Viabilidade de sementes; Reconhecimento de padrões; Árvore de decisão; Classificação supervisionada; Seed quality; Seed vigor; Seed viability; Pattern recognition; Decision tree; Supervised classification; CIENCIAS AGRARIAS::ENGENHARIA AGRICOLA

Production and use of high quality seeds are important factors for the soybean farming. Therefore the quality control system in the seed industry must be reliable, accurate and fast. Seed technology research has been striving to develop or improve tests to enable seed quality evaluation. Tetrazolium test, besides evaluating the viability and vigor of the seeds, provides information about the potencial causing agents of quality reduction. Even though not using expensive instruments and reagents, the test requires a well-trained seed analyst, and the test’s accuracy depends on their knowledge about the all involved techniques and procedures, including the subjectivity of the observer. Therefore, the objective of the present research was to develop a computational tool that could minimize the implicit subjectivity in the test, contributing to increase information credibility and ensure the accuracy results. This tool allows, by tetrazolium test images, to identify seeds damage, as well as their location and extension, making the interpretation less subjective. From the feature extraction data in digital images of tetrazolium test, supervised classification algorithms were applied to do segmentation in the images, generating a classified image. The proposed system was tested using a selection of samples to training the classifier model and, from this model, the images classification of the tetrazolium test, to extract information about the seeds damage. The system allowed, in addition to an easier way for damages identification in the tetrazolium test images, the extraction of accurate information on displayed damage and achieve the control of the analyzed samples. The classifier performed the assignment of the predetermined categories efficiently for non-present data training set, with 96.6% of correctly classified instances and Kappa index of 0.95%, making the system a supplementary tool in decision making for the tetrazolium test.