Detalhes bibliográficos
Ano de defesa: |
2023 |
Autor(a) principal: |
Zaions, Deividi Felipe
 |
Orientador(a): |
Hölbig, Carlos Amaral
 |
Banca de defesa: |
Não Informado pela instituição |
Tipo de documento: |
Dissertação
|
Tipo de acesso: |
Acesso aberto |
Idioma: |
por |
Instituição de defesa: |
Universidade de Passo Fundo
|
Programa de Pós-Graduação: |
Programa de Pós-Graduação em Computação Aplicada
|
Departamento: |
Instituto de Ciências Exatas e Geociências – ICEG
|
País: |
Brasil
|
Palavras-chave em Português: |
|
Área do conhecimento CNPq: |
|
Link de acesso: |
http://tede.upf.br:8080/jspui/handle/tede/2514
|
Resumo: |
Desonivalenol is one of the most commonly mycotoxins found in wheat grains due to Fusar- ium contamination. Its ingestion causes toxic effects, causing risk to both human and animal health, as well as the loss of economic value. Thus, the development of fast and robust tech- niques to the identification of this mycotoxin at the beginning of the production chain is of ut- most importance for global food security. Non-destructive techniques based on near-infrared (NIR) spectroscopy have shown promising prospects for the detection of DON contamination in wheat grains. However, recent studies use high-cost hyperspectral imaging equipment for data collection. As a result, complex mathematical models and high demanding processing are required to identify DON contamination. This work aims to develop a multispectral sys- tem for the identification of DON mycotoxin in wheat grains. Using multispectral sensors that operate within the visible and NIR spectrum, a portable system capable of communicating with a mobile device (smartphone) is designed. Two algorithms based on neural networks are then developed, one for regression and the other for binary decision between healthy and contaminated, and a mobile application to receive the spectral data from the sensor and process it together with the algorithms. For training, testing and validation of the proposed system it is used a total of 117 samples with different levels of contamination and 84 healthy samples. Through the obtained results it is checked that the regression algorithm did not show sufficient accuracy to indicate the levels of contamination. On the other hand, from the combination between the binary algorithm and the adaptation of the regression algorithm for binary decision, the system presented an accuracy R² of 0.855 for the identification of healthy and contaminated samples. With this, we could conclude that multispectral sen- sors operating in the visible and NIR region are capable of identifying DON contamination in wheat grains. |