Detection of Peruvian Golden Berry Quality using a Convolutional Neural Network

Taquila-Velásquez J.; Vela-Guerra A.; Portillo-Mendoza P.; Sotomayor-Beltran C. · International Journal of Engineering Trends and Technology (2024) · DOI: 10.14445/22315381/ijett-v72i1p129

Resumen

- In Peru, the National Strategic Export Plan seeks to promote exports of many fruits, among them the Peruvian golden berries. To increase the production speed of golden berries to match foreign competition, the design of a convolutional neural network (CNN) algorithm applied to the selection of this type of berry is proposed. The classification of the golden berry quality has been defined according to its condition, a good one and in poor condition. The developed classification software, at its core, is based on a pre-trained CNN called GoogLeNet, which was implemented using Matlab development software; additionally, the Matlab App designer extension was used to create a graphical user interface. The architecture of the CNN allowed us to obtain the characteristic parameters of the fruit, where the criterion for using the database considered 80% images for training, 10% for validation, and 10% for post-training. For this work, 241 images were used, divided into two blocks, as NO_GOOD (in poor condition and immature) and as GOOD (mature and in good condition). The best-trained CNN, with a validation percentage of 100%, was embedded in the user interface. A total of 6000 iterations and 500 epochs were used for the training of the CNN. The development of the user interface system allowed to select golden berries based on their quality at greater speeds than by doing it manually.

Instituciones: Universidad Tecnológica del Perú