Detalhes bibliográficos
Ano de defesa: |
2018 |
Autor(a) principal: |
Pierre Júnior, Mário Lúcio Gomes de Queiroz
 |
Orientador(a): |
Angelo, Michele Fúlvia
 |
Banca de defesa: |
Não Informado pela instituição |
Tipo de documento: |
Dissertação
|
Tipo de acesso: |
Acesso aberto |
Idioma: |
por |
Instituição de defesa: |
Universidade Estadual de Feira de Santana
|
Programa de Pós-Graduação: |
Mestrado em Computação Aplicada
|
Departamento: |
DEPARTAMENTO DE CIÊNCIAS EXATAS
|
País: |
Brasil
|
Palavras-chave em Português: |
|
Palavras-chave em Inglês: |
|
Área do conhecimento CNPq: |
|
Link de acesso: |
http://tede2.uefs.br:8080/handle/tede/1343
|
Resumo: |
Brazil is one of the main producers of orange juice, exporting 98% of its production.Phytosanitary problems cause loss of production and difficulty in exporting. Among existing pests is the industry’s biggest concern that Huanglongbing, also known as Greening. This disease reduces the quality of the fruit, disrupts the developmentof the plant and reduces its productivity. The main spreader of the disease-causing bacteria is the Diaphorina citri, which is 2 to 3 mm in length. For the control of the problem, one method is to capture from yellow traps and the count of the insects that insects to later adjust the dosage of the insecticides to be applied. This way of monitoring is an important component in the prevention, the detection and counting of the insect that causes the disease is carried out manually and this process is determinant for more effective insecticide applications. This research had as general objective the use of the methodology of Deep Learning with Neural Convolutional Networks (CNN) in the classification of the insect Diaphorina citri in digitized images of adhesive traps, in order to streamline the identification process and improve the accuracy results in the recognition of the insect. To do this, it was necessary to create a database with the images of the traps after digitized and to analyze the results obtained in the classification of Diaphorina citri using models of distinct architectures with deep learning approach. For automated classification, three architectures of Neural Convolutional Networks (CNN) were tried. After evaluation and application of statistical tests to compare the results of the LeNet, AlexNet, Inception architectures, the Inception model applied to the set of test samples for generalization of the model presented an average accuracy of 99.51% accuracy inthe crossvalidation and 99.37% in the validation with the final test set in the classification of the insect Diaphorina citri. |