Premium
Using Variability within Digital Images to Improve Tall Fescue Color Characterization
Author(s) -
Ghali Ihab E.,
Miller Grady L.,
Grabow Garry L.,
Huffman Rodney L.
Publication year - 2012
Publication title -
crop science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.76
H-Index - 147
eISSN - 1435-0653
pISSN - 0011-183X
DOI - 10.2135/cropsci2011.10.0553
Subject(s) - calibration , statistics , mean squared error , linear regression , artificial intelligence , festuca arundinacea , mathematics , coefficient of determination , computer science , biology , agronomy , poaceae
Digital image analysis (DIA) provides an accurate, nondestructive, and objective assessment of turf color. Previous research developed an index known as the dark green color index (DGCI) via DIA as an indicator of turf color. The objective of this study was to use DGCI variability to better predict a visual rating (VR) index used to evaluate tall fescue ( Festuca arundinacea Schreb.) color under different irrigation treatments. To develop DGCI statistics, two freeware software packages (Image J and R) were used to extract and process information from digital images. The model to predict VR from DIA was developed and calibrated using candidate DGCI statistical moments from 120 images in a calibration data set using a multiple linear regression procedure. Fitness of calibration and validation models were verified using the adjusted coefficient of determination, root mean square error, and the Mallow's C p statistic. The two‐variable model produced more precise estimates (adjusted R 2 = 0.926 and 0.899) than the model that only used one term in predicting the VR values (adjusted R 2 = 0.879 and 0.843) for calibration and validation sets, respectively. These data suggest incorporating a measure of color uniformity improves the use of DGCI in predicting VR values compared to using only the mean of DGCI values to predict VR values. Model refinements may be needed for other turf species, but current work suggests using additional statistical moments such as SD improves VR estimate precision and accuracy.