Image-based sensor for on-tree automatic color tracking in pomegranate orchards

Harvesting decisions based on crop monitoring are an emerging goal of precision agriculture techniques. In the case of pomegranate (Punica granatum L. ‘Wonderful’), color is a good indicator of fruit ripeness, and fruit ripening stage is used as a quality parameter. In this work, a low-cost autonomous sensor to perform continuous on-tree monitoring of the pomegranate color evolution was developed. The sensor measurement is based on the image captured by an RGB camera. To measure the fruit color, a Deep Learning based instance segmentation model was specifically generated for pomegranates. The model achieved an F1 score of 0.817. Nine sensors were installed in the field on fixed platforms and took measurements continuously for two seasons. The fruit color time curves demonstrated the utility of the proposed sensor to inform about the evolution of the crop ripening. The performance of the sensors was evaluated by comparing on several dates their measurements with those of a colorimeter on fruit samples in the laboratory. The results showed a very high correlation, with a mean R
²
of 0.94.