Field-based Robotic Phenotyping for Sorghum Biomass Yield Component Traits Characterization Using Stereo Vision
Sorghum is known as a major potential feedstock for biofuel production. Being able to efficiently discover genetic control of many traits over a large number of genotypes, genome-wide association study (GWAS) has become a powerful tool for studying sorghum biomass yield components. However, automated high-throughput field-based plant phenotyping is now the bottleneck for scaling up such experiments. This paper presents an auto-guidance enabled utility tractor which navigates itself between crop rows with a predefined path while collecting stereo images of sorghum samples from both sides of the vehicle. Three levels of stereo camera heads were instrumented to capture images of plants up to 3 meters tall. The stereo images were processed offline to reconstruct 3D point clouds using Semi-Global Block Matching. A semi-automated software interface was developed to measure stem diameter due to the strict sampling strategy and the complexity of high-density crop canopy. An automated hedge-based feature extraction pipeline was proposed to quantify other variations in plant architecture traits such as plant height, leaf area index (LAI) and vegetation volume index (VVI). The stem diameter measured using the semiautomatic method showed high correlation (0.958) to hand measurement.
This article is published as Bao, Yin, and Lie Tang. "Field-based robotic phenotyping for sorghum biomass yield component traits characterization using stereo vision." IFAC-PapersOnLine 49, no. 16 (2016): 265-270. DOI: 10.1016/j.ifacol.2016.10.049. Posted with permission.