Phenotypic Parameter Extraction System for Crops Based on Supervoxel Segmentation

Abstract

Obtaining crop structural parameter information is an important way to study crop's growth, development status, and accumulation biomass. Currently, the measurement of vegetable crop's phenotypic parameters is time-consuming and can cause damage to crops. Thus there is a demand for rapid non-destructive measurement of crop phenotypic parameters. In this paper, we design a system to extract the two main parameters (leaf area and average leaf angle). The initial point cloud obtained from an RGB-D camera is segmented by employing Locally Convex Connected Patches based on supervoxel clustering. After comparing with other reconstruction algorithms, we choose Greedy Projection Triangulation to reconstruct the segmented leaves. In addition, random sample consensus is used to extract phenotypic parameters from the constructed mesh. More than one hundred sets of RGB-D data are collected to verify the feasibility of the system. Experiments show that the system is able to segment most of the leaves effectively and the extracted phenotypic parameters achieve acceptable accuracy.