A smart system for the automatic evaluation of green olives visual quality in the field

• Published in: Computers and electronics in agriculture 2020-12, Vol.179, p.105858, Article 105858
• Main Authors: Sola-Guirado, Rafael R., Bayano-Tejero, Sergio, Aragón-Rodríguez, Fernando, Bernardi, Bruno, Benalia, Souraya, Castro-García, Sergio
• Format: Article
• Language: English
• Subjects: Bruising; Color; Computer vision; Decision making; Image acquisition; Image analysis; Image segmentation; Mathematical models; Maturity index; Olives; Parameters; Performance indices; Root-mean-square errors; Supervised training; Visual fields; Weight; Weight estimation
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2020.105858

•A system for the automatic analysis of olive parameters in-field has been developed.•The system is based on image making that allows the user to make technical decisions.•The system is low cost and ready to use, only requires taking fruit from the field.•Colour parameters have been determined to sort of the fruit maturity.•Supervised training models have been obtained for the fruit bruising quantification. Monitoring some of the parameters that affect the quality of table olives for green processing is fundamental in a farmer's decision making. This work develops an affordable system for in-the-field evaluation of fruit calibre, ripeness and bruise index. The system consists of an illuminated cube that acquires images of fruit samples and generates an instantaneous report, using computer vision techniques implemented in software. To do this, it was necessary to determine models of fruit weight and size and also the colour regions (RGB colour space) involved in olive maturity indexes. Moreover, supervised training models were created to perform image segmentation (background and bruising areas). Error in the estimation of fruit weight was very low (R2 = 0.9), and prediction of the maturity index (MI) was quite good, with an accuracy of 0.66 and 0.91 for manually sorted olives in MI0 and MI1 respectively (green processing). Prediction of MI2 had lower precision (0.48) when the fruit was changing to black-purple and the bruising spots were confused with fruit area because of determined similarities in colour. The error in the estimated bruise index was lower for MI0 (RMSE = 2.42) than for MI1 (RMSE = 3.78), both of which are suitable for an estimation of quality in the field. Overall, the system's performance reveals promising results for a quick, easy and accurate evaluation of the external parameters that define the quality of olives. The models obtained could be useful for other purposes.