Grasses within the Sporobolus genus have been classified as problematic weeds of pastures in many countries. In Florida, giant smutgrass is the most common and troublesome weedy Sporobolus grass. The use of unmanned aerial vehicles (UAVs) for mapping, combined with site-specific weed control has the potential to optimize giant smutgrass management and decrease the use of herbicides. In this research, RGB ortho-mosaics captured from a simple UAV were examined to detect and map giant smutgrass in bahiagrass pastures in Florida. Two sampling dates (May and August) and four flight altitudes (50, 75, 100 and 120 m) were investigated for optimal classification accuracy. Spectral, texture and combined (spectral and texture) analyses served as the basis for supervised (random forest) and unsupervised (k means) classifications. Giant smutgrass cover was successfully mapped and best evaluated by integrating the combined analysis with supervised algorithm, reaching a correlation of 0.91 with the ground truth cover. Flight altitude had a negative relationship with giant smutgrass detection; however, satisfactory results were also obtained from 120 m with an average correlation of 0.76 when using combined supervised classification. Additionally, both sampling dates were found adequate for giant smutgrass mapping. These findings demonstrate that low-cost UAV platforms can successfully be used to generate accurate giant smutgrass infestations maps, allowing for site-specific management in bahiagrass pastures. Results from this work also broaden the general knowledge on the impacts that different settings and parameters (e.g. time of the year, altitude and image-analyses methods) can have on aerial image classification.
Grasses within the Sporobolus genus have been classified as problematic weeds of pastures in many countries. In Florida, giant smutgrass is the most common and troublesome weedy Sporobolus grass. The use of unmanned aerial vehicles (UAVs) for mapping, combined with site-specific weed control has the potential to optimize giant smutgrass management and decrease the use of herbicides. In this research, RGB ortho-mosaics captured from a simple UAV were examined to detect and map giant smutgrass in bahiagrass pastures in Florida. Two sampling dates (May and August) and four flight altitudes (50, 75, 100 and 120 m) were investigated for optimal classification accuracy. Spectral, texture and combined (spectral and texture) analyses served as the basis for supervised (random forest) and unsupervised (k means) classifications. Giant smutgrass cover was successfully mapped and best evaluated by integrating the combined analysis with supervised algorithm, reaching a correlation of 0.91 with the ground truth cover. Flight altitude had a negative relationship with giant smutgrass detection; however, satisfactory results were also obtained from 120 m with an average correlation of 0.76 when using combined supervised classification. Additionally, both sampling dates were found adequate for giant smutgrass mapping. These findings demonstrate that low-cost UAV platforms can successfully be used to generate accurate giant smutgrass infestations maps, allowing for site-specific management in bahiagrass pastures. Results from this work also broaden the general knowledge on the impacts that different settings and parameters (e.g. time of the year, altitude and image-analyses methods) can have on aerial image classification.