Multi-phenotypic parameters extraction and biomass estimation for lettuce based on point clouds

Traditional phenotypic measurements were destructive, time-consuming and laborious. 3D computer vision technology collects plant information in a non-destructive manner and can work continuously. The purpose of this paper is to measure lettuce multi-phenotypic parameters and estimate biomass by point clouds. The result shows: (1) RGB information improved the segmentation effect of the point cloud. (2) The point cloud segmentation method proposed in this paper has a high accuracy (accuracy = 98.7 %) and keeps high spatial resolution. (3) On the test set, the R2 of height, diameter, leaf area, fresh weight and dry weight are 0.935, 0.905, 0.969, 0.966 and 0.968, respectively, the cumulative NMSE is as low as 0.069. The Totalscore is 15 % better than the 1st of the online challenge held based on this data set. The results showed that the plant phenotypic extraction method proposed in this paper was accurate and efficient, and could be applied for high-throughput phenotyping.