Sistema inteligente de selección de cosecha en cultivo de pitahaya mediante visión artificial

Abstract

Quality control of export fruits in Ecuador, such as yellow pitahaya, is essential because it is widely accepted in both domestic and international markets. The technical standard NTE INEN 025 establishes criteria for the classification of fruit by degrees of ripeness for export and domestic consumption. The objective of this research work is the development of an artificial vision system for crop selection in pitahaya cultivation that allows the farmer to automatically select this fruit for export, avoiding human error. This system involves the use of the YOLOv5 model for deep learning in the agricultural context, which evaluates the ripening of pitahayas in three stages. The prototype is based on a cabin composed of primary elements for artificial vision, which is a high-resolution webcam and artificial white light. The system captures the image of six fruits and classifies them according to the degree of ripeness of the pitahaya in an estimated time of 4.48 seconds. A graphical interface programmed in Python is used for data processing. Tests were carried out to evaluate the functionality and verification of the prototype, obtaining a reliability level of 91.82%. These tests involved various pitahayas at different degrees of ripeness, which leads to the conclusion that the system is useful for farmers or persons in charge of grading yellow pitahayas destined for both export and domestic consumption.