Development of an artificial vision system with convolutional neural networks (YOLOv8) for quality management of avocados

León León R.A.; Espejo Pócito B.R.; Guzmán Córdova E.F.; Huamanchumo Cortez F.A.; Morales Rodriguez H.A.; Nuñez Flores P.C. · CISCI - Conferencia Iberoamericana en Sistemas, Cibernetica e Informatica (2024) · DOI: 10.54808/cisci2024.01.461

Resumen

The project focuses on the development of an artificial vision system based on convolutional neural networks (YOLOv8) for the quality management of avocados. The main objectives are to improve accuracy and efficiency in avocado inspection, optimize selection and reduce product waste. 700 images of avocados captured with a high-resolution camera were used, varying lighting conditions and angles. They were labeled according to their maturity in Roboflow, and data augmentation techniques were applied. The YOLOv8 model was trained on Google Colab with a Tesla T4 GPU, adjusting parameters such as batch size and learning rate for 150 cycles. It was evaluated with a test data set and a Python algorithm was implemented in PyCharm to evaluate the maturity of the avocados, using Google Drive to store data and models. The results obtained show that the YOLOv8 model has high accuracy in avocado classification. Model accuracy ranged from 86% to 95% for the “good” category and 86% to 90% for the “bad” category. © CISCI - Simposium Iberoamericano en Educacion, Cibernetica e Informatica 2024.

Instituciones: Universidad Privada del Norte