Development of an Automated System with Convolutional Neural Networks (CNN) for the Detection of Black Spots in Mangoes

León León R.A.; Estela Medina N.M.; Lucano Tolentino M.T.; Teodor Soriano J.E.; Padilla Flores J.M.; Herrera Ortiz E.E. · CISCI - Conferencia Iberoamericana en Sistemas, Cibernetica e Informatica (2025) · DOI: 10.54808/cisci2025.01.344

Resumen

The study aimed to develop an automated system based on convolutional neural networks (CNNs) for the detection of black spots on mangoes, with the aim of optimizing the fruit classification process in the field or at the collection center. A labeling dataset composed of images of mangoes with and without blemishes was used, training the model for 40 epochs. During training, the model adjusted its parameters to maximize its performance in binary classification tasks. At the end of the process, it achieved an accuracy of 96.82%, a recall of 95.58%, and an F1-score of 96.19%, while the mAP@0.5 was 99.14%. The F1 curve showed stable behavior in the last iterations, reflecting an optimal balance between precision and sensitivity. These results demonstrate the model's ability to reliably identify blemishes, even with variability in visual conditions. It is concluded that the proposed system is technically feasible for implementation in machine vision systems, bringing efficiency and objectivity to the selection process of mangoes for the market. Future improvements could focus on expanding the detection range and optimizing performance under conditions of greater visual complexity.

Instituciones: Universidad Privada del Norte