Green-gradient based canopy segmentation: A multipurpose image mining model with potential use in crop phenotyping and canopy studies

Abstract

Light extinction is the most fundamental aspect of green canopies. The exponential form of light gradient is extensively evaluated or utilized by conventional approaches mainly in contribution to the vital concept of leaf area index (LAI), which reasonably characterizes canopies based on their theoretical capability for light attenuation i.e. having greater leaf surfaces. We analyzed the image archive of heterogeneous wheat canopies (cultivar mixtures), captured from experimental plots of a two-year field study conducted at College of Agriculture, Shiraz University, Iran, to evaluate the option of using commercial digital cameras for comparing the canopies based on their relative optics, and also providing new representation of light extinction within the canopy. Here, two novel distinct techniques including different imaging and calculation methods were employed. In the first one, or Green-based Segmentation Model (GSM), pixels of vegetation parts were categorized into "n" light-to-darkness groups based on their similar Green values (RGB color system). Then, mean red, mean green, and mean blue values of each group were calculated and their separate trends were plotted against the mentioned green light-to-dark steps. In the second method, a kind of in-situ spectroscopy of sun-exposed leaves (DLBE, Double-shot Light Balance Equations) was evaluated under the isotropic scattering assumption, with the aim of estimating the absorbed portion of light by sun-exposed leaves, using reflected and transmitted portions. Facing the sun, and then with the sun behind the camera, two images not peer-reviewed) is the author/funder. All rights reserved. No reuse allowed without permission.The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/241786 doi: bioRxiv preprint first posted online Jan. 1, 2018; 2 were taken from each plot; in which "Transmission" and "Reflection" ROIs (ranges of interest) were extracted, respectively. Finally, normalized color values were calculated using mean RGBs of either type of ROI. Results of GSM showed that the green-scaled gradient of reflected red and blue lights had robust exponential trends which could be readily utilized for identifying canopies in the form of one or two image-derived equation (