Qualidade fisiológica de sementes de gergelim por meio do processamento digital de imagens

• Ano de defesa: 2024
• Autor(a) principal: Pinheiro, Paloma Rayane
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Sesamum indicum L.; ImageJ; Máquina de vetor de suporte; Floresta aleatória; K-vizinhos mais próximos; Gergelim - Semente; Gergelim - Sementes - Qualidade; Sesamum indicum L; Support vector machine; Random forest; K-nearest neighbors; Sesame - Seeds - Quality; CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::FITOTECNIA
• Link de acesso: http://repositorio.ufc.br/handle/riufc/79529

Sesame (Sesamum indicum L.), an oilseed widely cultivated worldwide, is used in a variety of sectors, from cooking to medicine. With the increasing demand for high-quality seeds, there is a growing search for tools that offer new data and correlate with traditional vigor tests, with faster and more efficient analyses. In this context, innovative methodologies, such as image analysis through the use of software and machine learning, have gained prominence. Given this scenario, two main objectives were established: 1 - Determine the vigor of sesame seeds through images of seedlings using the ImageJ software. 2 - Classify sesame seeds based on their vigor, through the analysis of digital images using machine learning, in addition to comparing the efficiency of three classifiers (SVM, KNN and RF). In experiment 1, five batches of sesame seeds were used. In addition to performing traditional tests (PC, G, IVG, MS, EA and E), the seeds were scanned and then taken to the germinator. The resulting seedlings were photographed and analyzed using ImageJ software, which allowed the measurement of root and shoot length. The images were taken at two time intervals (three and six days) and under two temperature conditions (30 °C and 35 °C). The results obtained from ImageJ were used to calculate new variables, processed by the SeedCalc plugin in the R® program. In experiment 2, sesame seeds were scanned, where a total of 426 images were generated, previously classified into two vigor categories (high and low), based on the length of the seedlings. Subsequently, these images were analyzed using machine learning techniques. The results were evaluated using a confusion matrix, to obtain the calculation of metrics, such as accuracy, precision, recall and F1-score. In experiment 1, a correlation was established between the results of the seedlings analyzed by ImageJ and vigor indices obtained by traditional tests. It was observed that the six-day period, at both temperatures evaluated, presented the best results. The analysis of images of the seedlings proved to be an efficient tool for evaluating the quality of sesame seeds, and the variables obtained were adequate to classify the lots satisfactorily. In experiment 2, the machine learning-based classifiers demonstrated to differentiate the sesame seeds in the two established vigor categories. Among the models evaluated, the SVM and RF algorithms obtained the best results.