Influence of Computer Vision for prediction of harvest in high stem fruits: Systematic review

Augusto Z.S.C.; Lozada J.F.P.; Yasmín G.P.L.; Doriana Hesseler T.D. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2024) · DOI: 10.18687/leird2024.1.1.645

Resumen

The accelerated development of computer vision globally is having a significant impact on fruit harvest estimation, leading to increased efficiency and significant reduction in food waste. Furthermore, this technology faces notable resistance and lack of knowledge in the agricultural sector. The objective of this review is to analyze artificial vision methods in predicting the harvest of high-stemmed fruits. Therefore, a non-experimental descriptive design was applied, belonging to a systematic review without meta-analysis. Based on the defined criteria (inclusion and exclusion), 26 open access articles were chosen from the electronic databases Scopus, Scielo and Redalyc, which address the use of VA to predict the harvest of high-stemmed fruits. The findings indicated that, most studies used near-infrared (NIR) spectroscopy and RGB image processing to estimate harvest, achieving average accuracies of 95% (citrus fruits) and 75% (apples) respectively. Furthermore, the use of UAVs with RGB and YoloV3 image sensors made it possible to achieve accuracies greater than 90%, enabling predictions made between 4 and 6 months before the harvest. It was concluded that the most used VA methods were RGB image sensor, spectrometry (NIR), unmanned aerial vehicles (UAV) and YOLOv3, which demonstrated accuracies greater than 75% in predicting the maturity of high-stemmed fruits. The choice of the method will depend mainly on whether you want to analyze the internal or external part of the fruit or both, therefore, it is important to recognize the variation in the pigmentation of the shell of the high-stemmed fruit in its growth stages.

Instituciones: Universidad Tecnológica del Perú