Banca de QUALIFICAÇÃO: JOAO CELIO LUNA DE CARVALHO

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : JOAO CELIO LUNA DE CARVALHO
DATE: 25/06/2026
TIME: 13:00
LOCAL: Google meet
TITLE:

APPLICATION OF BIOSPECKLE LASER ASSOCIATED WITH CONVOLUTIONAL NEURAL NETWORKS FOR MONITORING THE QUALITY OF TOMATO FRUITS

KEY WORDS:

Image processing, machine learning, posthaverst.

PAGES: 29
BIG AREA: Ciências Agrárias
AREA: Engenharia Agrícola
SUMMARY:

Biospeckle Laser (BSL) is a non-destructive optical technique associated with the activity of biological materials that has shown great success in applications related to food quality. BSL patterns are obtained over a specific period of time, composed of a set of frames (speckles), which can be analyzed in their original form or transformed into Time History of the Speckle Pattern (THSP) matrices, condensing the information into a single image. Although traditional methods for analyzing BSL patterns are well-established, relevant information can be lost when focusing on extracting specific features. Convolutional Neural Networks (CNNs) are deep neural networks capable of autonomously extracting relevant features, specializing in the analysis of two-dimensional data (2D CNNs). With the need to evaluate data over time, CNNs for three-dimensional data (3D CNNs) emerged. There are few studies associating CNNs and BSL, and no records of THSP data analysis with 2D CNNs. This approach can bring benefits such as lower computational cost and faster data processing, being applicable in optimizing the monitoring of agricultural product quality, such as monitoring the tomato drying process. Thus, this project aims to propose a new approach for BSL analysis based on the classification of THSP matrices using CNNs and evaluate its application for monitoring the drying process of tomato fruits. To this end, two computational routines will be developed: one for the construction and quantitative analysis of THSP matrices, and another for qualitative analysis of BSL patterns and THSP matrices using 3D and 2D CNNs, respectively. The developed software will be tested with simulated BSL data, called synthetic data, generated by a third computational routine. Subsequently, existing data from previous research corresponding to seed germination vigor and internal bruising in potatoes will be used. Finally, data on tomato fruit drying in a laboratory oven will be collected. The obtained BSL patterns will be quantitatively characterized using absolute difference (AVD) and moment of inertia (MI) values, and the classification results obtained from training the 2D and 3D CNNs will be evaluated using performance metrics (accuracy, precision, recall, and F1-score) and processing time. The project aims to produce a software suite that analyzes BSL patterns and demonstrates the feasibility of using THSP matrices in 2D CNNs for pattern classification.

COMMITTEE MEMBERS:
Interno - 1050615 - MURILO MACHADO DE BARROS
Externo à Instituição - ADILSON MACHADO ENES - UFS