Grassland Species Identification and Mapping with UAS Imaging

Identifying grassland species and estimating their coverage is crucial for effective forage management and biodiversity monitoring. There is significant promise in leveraging multispectral drone imagery and machine learning algorithms, while the potential for multi-species identification from drone imagery is still unclear. Here, we aim to investigate which grassland species can be accurately identified and to assess the impact of time and location to classification accuracy. During the experiment drone images were capture and several plant species of interest identified, particularly those relevant to forage quality and biodiversity issues. Images were processed further and utilised to generate a synthetic dataset for training neural networks on a semantic segmentation task to recognise these species in drone imagery throughout the vegetative period. Results have demonstrated significant potential for identifying common species and even distinguishing between various grass species with the help of additional infrared bands in multispectral imagery. The incorporation of multi-temporal analyses has enhanced classification accuracy, especially in areas with mixed species. Multi-species detection in grassland seems possible and will be enhanced by further model training and continuous learning processes. Moreover, robust classification models have the potential to improve grassland management strategies and contribute to more effective biodiversity monitoring methods and conservation efforts.