Deep Learning-Based System for Pineapple Eye Scanning and Extraction Using YOLOv8n and GRBL

Franco R.; Gonzales J.; Nieves A.; Carrizales J.; Figueroa J.; Sobrado E. · 2026 The 9th International Conference on Image and Graphics Processing, ICIGP 2026 (2026) · DOI: 10.1109/icigp68997.2026.11619721

Resumen

Pineapple processing still relies on manual "eye"removal, which requires skilled labor and often causes avoidable pulp losses due to inconsistent positioning and cutting depth. This paper presents a low-cost prototype focused on (i) scanning/detecting pineapple eyes using a single RGB view and (ii) executing extraction motions through GRBL-controlled actuation. The perception module uses YOLOv8n trained for one class (pineapple eye), and the motion module maps detections to a commanded rotation coordinate and tool approach-drill-retract trajectories.Two training configurations were evaluated during development: a dataset of 200 images (selected final model) and another of 127 images (baseline reference), captured from approximately 20 pineapples under controlled conditions. The camera was positioned at a fixed distance of 16 cm. The current system employs a single-view strategy and an empirical mapping between GRBL Y-axis command and angular rotation, θ=Y/40. Results show that increasing the dataset size improves training stability and reduces misdetections. Current limitations include calibration refinement, illumination sensitivity, pineapple size variability, and the single-view constraint.

Instituciones: Universidad Peruana de Ciencias Aplicadas