Measurement of the ripening rate on coffee branches by using 3D images in outdoor environments

In this article, a method for determination of the ripening rate of coffee branches is presented. This is achieved through analysis of 3D information obtained with a monocular camera in outdoor environments and under uncontrolled lighting, contrast, and occlusion conditions. This study was performed on 30 coffee branches, monitored throughout the entire harvest season, at each of nine collection times, directly in the countryside. On each branch checked, the fruit was counted in three maturation stages: unripe, semi-ripe and ripe. Subsequently, the real ripeness percentage was computed for each stage of development. Images were acquired from the same manually checked branches, and the developed algorithm was run. This algorithm performs a reconstruction producing a point cloud, employing Structure from Motion (SfM). In all, 17 features were selected to describe the three maturation stages, leaves, and stem. These features were calculated in a region proportional to the size of the point cloud. The 3D structures of the coffee fruits on the branch were obtained and classified, with between 42% and 92% of correct classification percentage for the unripe, semi-ripe, and ripe developmental stages, and the best classification model was selected, with a neighborhood that was 6.5% of the total size of the point cloud. A statistical estimation model was constructed for each stage of development, and it was found that the model overestimates the percentage of ripe fruit, underestimates that of semi-ripe fruit, and the unripe fruit estimate is one to one. The errors in estimation were between 3.7% and 8.7%. Correlation was between 0.64 and 0.98. Finally, temporary monitoring was performed for the ripening of coffee branches during the harvest season, with nine collection times. A maturation index was determined, with which the state of a branch as ready or not for harvest was correctly determined with 83% efficiency.