Artificial Intelligence in the Optimization of Macauba (Acrocomia aculeata) Processing: Recognition of Phytobiometric Patterns

This study assessed the YOLO model’s effectiveness in identifying the phytobiometric patterns of Macaúba (Acrocomia aculeata), specifically detecting and classifying immature fruits. Conducted at Fazenda Limeira, Lavras, Minas Gerais, Brazil, the experiment used Macaúba fruits from spontaneously grown palm trees in a grazing area. Fruits were randomly selected, cleaned, and stored for 48 hours in a shaded area before photographic recording and biometric measurement. Performance analysis showed significant improvements, highlighting the model’s potential to optimize agro-industrial processes. Key loss metrics were substantially reduced: box_loss from 0.9 to 0.2, cls_loss from 1.4 to 0.1, and dfl_loss from 1.4 to 0.8. These reductions indicate enhanced accuracy in localization and classification, demonstrating YOLO’s ability to adjust bounding boxes and correctly identify fruit classes. Furthermore, the model achieved perfect Precision and Recall values (1.0) in the precision-recall (PR) curve, eliminating false positives and negatives. This ensures precise matching between predictions and actual instances of the Immature Macaúba ("Macuba Imatura") class. Applying the model in Macaúba processing offers benefits such as automatic sorting, improved quality control, and waste reduction. This work emphasizes the feasibility of using artificial intelligence in precision agriculture, promoting automation and sustainability within production chains.