Deep learning-based quality control system for meat products

Abstract

The meat industry faces significant contamination issues that have led to disease outbreaks and high levels of food waste, with losses of up to 98,550 tons per year in Peru. This study presents a system based on Convolutional Neural Networks (CNNs), using a RESTful API for the quality control of beef products at a distribution center in Lima, Peru. A dataset of 2,313 images was collected and categorized into four beef quality levels: first, second, third, and industrial. Of these, 1,849 were used for the model training phase, 224 for the validation phase, and 240 for the experimental phase. During the model validation phase, conducted using the validation set, the CNN achieved a validation accuracy of 75.4%, with a sensitivity of 75.0%, a specificity of 73.4%, and an F1 score of 0.73. This validation accuracy provided an estimate of the model’s generalization capability prior to final deployment. Subsequently, a real-world experiment was performed using the test set, simulating operational conditions with data from a meat trading company, to evaluate the deployed system’s performance in a practical setting. It achieved a test accuracy of 97.9%, demonstrating its robustness and applicability in improving quality standards for automated meat classification processes.