Software design for image mapping and analytics for high throughput phenotyping

• Published in: Computers and electronics in agriculture 2021-12, Vol.191, p.106550, Article 106550
• Main Author: Kim, James Y.
• Format: Article
• Language: English
• Subjects: Algorithms; Batch processes; Batch processing; Calibration; Cost analysis; Geofences; Image analysis; Image processing; Image segmentation; Mapping; Mathematical analysis; Micro air vehicles (MAV); Open source software; Phenotyping; Quantitative genetics; Remote sensing; Scientific visualization; Software; Soybeans; Spectral signatures; Spectrum analysis; Unmanned aerial vehicles
• ISSN: 0168-1699; 1872-7107
• DOI: 10.1016/j.compag.2021.106550

•User-friendly GUI-based software for HTP data analysis.•GIS interface for the user to interchangeably create shapefiles and a grid on the image canvas.•Geometric and radiometric calibrations.•Gridding method to align with field layout for plot-level image analysis.•High throughput image processing of aerial images for plot-level phenotypic metrics. Phenotyping is essential for the advancement of plant breeding and quantitative genetics. Image-based high throughput phenotyping (HTP) is a game changer to expedite a breeding pipeline by supporting large coverage and image analysis for morphological and spectral signature of the plant canopy. To meet the demand of a cost-effective and globally consistent HTP solution, a customized analytic tool, Image Mapping & Analytics for Phenotyping (IMAP), was developed to implement high throughput image processing and deliver plot-level metrics of plant phenotypes. IMAP is open source software written in Python and designed to provide data visualization and batch processes through algorithms for GIS interface, geometric and radiometric calibrations, geo-fencing, segmentation, and gridding. IMAP was applied and validated on aerial images collected by an unmanned aerial vehicle (UAV) and a manned aerial vehicle (MAV) on soybean field that was prepared for a drought resistance study under two different water treatments. The gridding algorithm extracted field boundary and delivered plot-level spectral analysis from aerial images in 1–4 s depending on the image size and the number of plots. The high correlation (R2 = 0.90 in vegetation and 0.89 in leaf area) between UAV and MAV images indicated the consistent performance of IMAP software in calculating the plot-level metrics and proved successful deployment of a high throughput phenotyping pipeline through a series of image processing and gridding algorithm in a batch process.