HyperPRI: A Dataset of Hyperspectral Images for Underground Plant Root Study

• DOI: 10.1101/2023.09.29.559614
• Publication Date: 2023-10-01T00:15:15Z
• Authors: Spencer J. Chang; Ritesh Chowdhry; Yangyang Song; Tomas Mejia; Anna Hampton; Shelby Kucharski; TM Sazzad; Yuxuan Zhang; Sanjeev J. Koppal; Chris H. Wilson; Stefan Gerber; Barry Tillman; Marcio F. R. Resende; William M. Hammond; Alina Zare
• Keywords: RGB color model; Phenomics

Abstract Collecting and analyzing hyperspectral imagery (HSI) of plant roots over time can enhance our understanding their function, responses to environmental factors, turnover, relationship with the rhizosphere. Current belowground red-green-blue (RGB) root imaging studies infer such functions from physical properties like length, volume, surface area. HSI provides a more complete spectral perspective plants by capturing high-resolution signature parts, which have extended beyond include physiological properties, chemical composition, phytopathology. Understanding crop plants’ physical, physiological, enables researchers determine high-yielding, drought-resilient genotypes that withstand climate changes sustain future population needs. However, most use cameras positioned above ground, thus, similar advances are urgently needed. One reason for sparsity is features often limited distinguishing reflectance intensities compared surrounding soil, potentially rendering conventional image analysis methods ineffective. Here we present HyperPRI, novel dataset containing RGB data in situ, non-destructive, underground using ML tools. HyperPRI contains images grown rhizoboxes two annual species – peanut ( Arachis hypogaea ) sweet corn Zea mays ). Drought conditions simulated once, boxes imaged weighed on select days across months. Along images, provide hand-labeled semantic masks environment metadata. Additionally, baselines segmentation this draw comparisons between focus spatial, spectral, spatialspectral predict pixel-wise labels. Results demonstrate combining HyperPRI’s spatial information improves target objects.