Diseño de un sistema de control de calidad con redes neuronales para clasificar guineo según su estado.

Abstract

Manual classification of bananas (Musa Cavendish) is prone to errors and subjectivity, leading to economic losses in the agroindustry. This paper presents a quality control system based on convolutional neural networks for classifying bananas according to their condition through computer vision. The system distinguishes between two categories: “Good” and “Bad.” A ResNet18 convolutional neural network with transfer learning was used, trained with images photographed in a controlled environment. The model was integrated into a desktop application developed with OpenCV and Tkinter, enabling real-time classification using a webcam and incorporating HSV pre- filters to improve detection. The evaluation metrics showed an accuracy level above 90%, thus validating its performance. The system reduces human intervention, improves quality inspection, and serves as an effective prototype for future larger-scale implementations in the food industry.