Evaluation of sensing and machine vision techniques in stress detection and quality evaluation of turfgrass species

Several field and greenhouse experiments were conducted to study the utility of image sensing and machine vision in stress detection and quality evaluation of turfgrass species. Results from the first phase of experiments indicated significant effects of ambient illumination correction, solar zenith correction and image segmentation on the reflectance indices values when assessing nitrogen status of creeping bentgrass and perennial ryegrass plots. In the second phase of research, greenhouse experiments were conducted from 2003 to 2006 to study the effects of nitrogen and water stress on reflectance characteristics of creeping bentgrass. Although water based index (WBI, 970 nm/900 nm) showed potential in differentiating N-stressed plants from water-stressed plants, further validation is necessary with other turfgrass species to conclude the possible utility of short wave infrared reflectance in differentiating N and water stresses in turfgrass.; The utility of different image sensing and non-image sensing techniques was also studied in objectively evaluating turfgrass quality parameters like, color, density and texture from National Turfgrass Evaluation Program trials. Image sensing took the greatest amount of time, while non-image sensing techniques were the fastest among the evaluated methods. All methods showed significant differences in cultivars for color in different trials. The quantified hue values from the multispectral camera were the least correlated with other evaluation techniques. Both chlorophyll meter and turf color meter showed potential in quantifying turfgrass color with greater consistency. However, the narrow separation obtained using turf color meter may not allow cultivar differentiation from species with less genetic color variation. Texture evaluation of turfgrasses was done after developing and implementing the run length encoding algorithm (RLE) on simulated turf built using twist ties in both planar and turf-type arrangements. Significant relationship was observed between manual measurements of twist ties and RLE-derived values. The algorithm implementation on true turfgrass images collected under greenhouse and field conditions from Kentucky bluegrass showed significantly positive relationship between RLE values and visual evaluation ratings. The possibility of collecting and analyzing images from multiple plots for color quantification was also evaluated successfully using an elevated platform from both Kentucky bluegrass and fairway bentgrass trials.