Banca de QUALIFICAÇÃO: JOÃO CÉLIO LUNA DE CARVALHO

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : JOÃO CÉLIO LUNA DE CARVALHO
DATE: 14/06/2024
TIME: 09:30
LOCAL: Google meet
TITLE:

Application of Convolutional Neural Networks to classify fruits based on quality attributes

KEY WORDS:

Image processing, machine learning, computer vision, tomato ripening, orange injuries.

PAGES: 30
BIG AREA: Ciências Agrárias
AREA: Engenharia Agrícola
SUMMARY:

The demand for quality horticultural products, correctly classified and selected, increases as food production and the world population grows. To assist in this activity, the application of artificial intelligence, with neural networks and image processing, has become increasingly common. Among the main crops with the highest production, commercialization and national consumption are tomatoes and oranges. The destination given by the industry to these fruits depends on quality standards, and equipment focused on the classification and selection of these fruits is still scarce in the national territory. Convolutional Neural Networks (CNN) are a subdivision of deep neural networks, dedicated to processing images in a faster and more efficient way. Therefore, the objective of this project is to use a computer vision system associated with Convolutional Neural Networks (CNN) to classify tomato and orange fruits based on quality attributes. Tomato fruits will be classified according to their state of maturity, while orange fruits will be classified according to the presence of injuries (defects). Both classifications will follow those proposed by the Companhia de Entrepostos e Armazéns Gerais de São Paulo (CEAGESP). 150 tomato fruits and 100 orange fruits will be used. The image acquisition system will consist of a Raspberry Pi 4 8 Gb coupled to a low-cost structure, along with its own camera module, an LCD monitor and halogen lamps for lighting. The distance between the camera module and the fruit will be standardized, and four images of each fruit will be captured in random positions, totaling 1000 images. With the database formed, two CNNs with defined architectures will be used, namely AlexNet and Mobile Net, trained without previously distributed prisoners, and the performances will be compared with a CNN of our own development, built during the project, entitled OakMoonNet. Performance comparison will be carried out by evaluating the Accuracy, Precision, Sensitivity (Recall) and f1-score parameters of each neural network used. Thus, with this project, it is expected to develop a Convolutional Neural Network with an efficiency comparable to existing CNNs, being lighter and with a shorter processing time. It is also expected to obtain a computer vision system from models based on convolutional neural networks, capable of classifying tomato fruits according to their degree of ripeness and orange fruits depending on the presence of injuries on the epicarp.

COMMITTEE MEMBERS:
Presidente - 2161955 - ANDERSON GOMIDE COSTA
Interno - 1050615 - MURILO MACHADO DE BARROS
Externo à Instituição - ROBERTO ALVES BRAGA JR - UFLA