GPhenoVision: A Ground Mobile System with Multi-modal Imaging for Field-based High Throughput Phenotyping of Cotton

Abstract. Phenotyping is an essential part of a new 'Green Revolution' to further improve crop productivity and quality. However, current field-based high throughput phenotyping (FB-HTP) systems have not incorporated multiple high-resolution imaging sensors, which limits their capability for measuring complex traits. The overall goal of this study was to develop and evaluate a modular and customizable FB-HTP system based on multiple imaging modalities to enhance the system extensibility to high data volume sensors and the capability to extract more traits. The system was developed upon a high-clearance tractor and contained sensing and electrical systems. The sensing system was based on a distributed structure, integrating environmental sensors, real-time kinematic GPS, and multiple imaging sensors including RGB-D, thermal, and hyperspectral cameras. Custom software with a multilayered architecture was developed using LabVIEW for system control and data collection. Validation and calibration were conducted for the three cameras. All sensors demonstrated acceptable accuracy and repeatability in measurements. The system was evaluated by scanning a cotton field with 100 plants, sampling 23 genotypes for quantification of canopy growth and development. Plant height, width, projected leaf area, and canopy volume and temperature were extracted from RGB-D and thermal images, and trait growth rates were calculated accordingly. Correlation results showed that the extracted traits had strong correlations (r-value ranged from 0.54 to 0.74) with fiber yield, suggesting the potential for establishing a yield prediction model useful for selecting high-yielding genotypes in cotton breeding programs, or assessing on-farm yield ahead of harvest. Combining growth rates of multi-dimensional morphological traits demonstrated the possibility of studying plant energy use efficiency in different growth stages and potential relationships with fiber yield. In addition, the extracted traits had a broad sense heritability (H²) ranging from 0.3 to 0.8, suggesting a great potential for quantitative genetic analysis. The fledgling system could be a useful tool for a wide range of breeding/genetic, agronomic/physiological, and economic studies.