A Computer Vision and Supervised Learning System for the Automatic Sorting of Persian Lime According to NMX-FF-077

The sorting of Persian lime Citrus × latifolia (Yu.Tanaka) Tanaka intended for export is still performed manually in many production units, introducing variability and low repeatability. Deep learning-based vision systems offer high accuracy but at a high cost, and with decisions that are difficult to trace against a quality standard. A system was developed that integrates classical computer vision (grayscale conversion, Gaussian filtering, thresholding and edge detection) with three supervised symbolic classifiers (PRISM, ID3 and Naive Bayes) under a hierarchical decision scheme, validated against the criteria of the Mexican Standard NMX-FF-077-1996-SCFI. The models were evaluated using a set of 7017 images, with a 265-image test subset, and the physical prototype was validated with 200 fruits. The system reached an accuracy of 95.1% on the test set and 95.5% during physical operation, with an F1 score of 0.97 for the export-grade class; only 2 of 265 and 1 of 200 non-conforming fruits were wrongly admitted. The cost of the deployed prototype remained at 407 USD. Integrating classical vision with interpretable symbolic rules constitutes an accessible and auditable solution for Persian lime quality control in accordance with the standard, with reproducible performance between algorithmic evaluation and physical operation.