Identification and Localization of Optimal Picking Point for Truss Tomato Based on Mask R-CNN and Depth Threshold Segmentation

Aiming at the problems of variable illumination, different fruit postures and the interference of main stem and fruit stem in natural greenhouse environment, this paper presents a new identification and localization method for picking point of truss tomato based on the combination of Mask R-CNN and depth threshold segmentation. 679 tomato clusters were detected in greenhouse. The results showed that the overall segmentation and detection success rate of Mask R-CNN ripe truss tomatoes reached 80.5%. On the basis of successfully obtaining the optimal picking point area, Mask R-CNN excluded the main stem and background fruit stem after depth threshold processing. The success rate of segmentation and recognition is 100%, and the success rate of fruit stem depth information positioning is 87.3%. The average detection time is 2.34s, which can be used to localize the optimal picking point in greenhouse.