Mask R-CNN for quality control of table olives
Miguel Macías Macías, Héctor Sánchez, Carlos J. Garcı́a-Orellana, Horacio M. González–Velasco, Ramón Gallardo–Caballero, Antonio García‐Manso
Abstract
Abstract In this paper we propose an object detector based on deep learning for scanning samples of table olives. For the construction of the system we have used a Mask R-CNN neural network. This network is able to segment the image providing a mask for each of the olives in the sample from which we can obtain the calibre of the object. In addition, the system is able to measure the degree of ripeness of the olives classifying them as green, semi-ripe and ripe, and identifying those fruits that are defective due to disease or damage caused by the harvesting process. The proposed system achieves success rates of 99.8% in the detection of olive fruits in photograms, 93.5% in the classification of fruit by ripeness and close to 80% in the detection of defects.