Comparison of grassland vegetation coverage extraction algorithms from UAV technology

Vegetation coverage is an objective index and an important parameter that reflects the basic extent of vegetation. In this study, we evaluated and analyzed the effectiveness of eight different visible-light-based vegetation indices for estimating the vegetation cover of different grassland types. Upon comparison of accuracy of these vegetation indices, we found that these vegetation indices were less effective in estimating the vegetation cover of desert grasslands. Therefore, we have proposed a desert vegetation index (DVI) to estimate the vegetation cover of desert grasslands. The effects of different vegetation indices on the vegetation cover estimation of different grassland types were evaluated, and the changes in the threshold values of different grassland types were analyzed. The results showed that: 1) The common vegetation indices could estimate the vegetation coverage in meadow grasslands and typical grasslands with a high accuracy (accuracy > 90%, F₁ > 0.9). The ExG (excess green index) was the best in estimating the vegetation coverage in meadow grasslands (accuracy > 93%, F₁ > 0.95), and there was no significant difference among the estimation power of vegetation indices when estimating the coverage in typical grasslands. However, the common vegetation indices exhibited low accuracy of vegetation coverage estimation (F₁ ≤ 0.6) in desert grasslands. 2) The DVI proposed in this study has a high estimation accuracy of vegetation coverage in desert grasslands (accuracy > 93%, F₁ score reached 0.71), which can effectively compensate for the defects of the above-mentioned vegetation indices. 3) The thresholds of GLI (green leaf index) and CIVE (color index of vegetation extraction) were the least sensitive to grassland types; the thresholds of ExG, ExGR (excess green minus excess red index), VEG (vegetative index), and WI (woebbecke index) were less sensitive to the effects of meadow grasslands and typical grasslands, but more sensitive to the effect of desert grasslands; and COM (combination index) and Lab (lab index) were the most sensitive to all the grassland types considered in this study.