Spectral analysis and mapping of blackgrass weed by leveraging machine learning and UAV multispectral imagery

• TGI composed of Green-NIR is the best spectral index for blackgrass weed. Random forest applied weed mapping with an accuracy 93%. Both and spatial information used classification. Algorithms are validated by UAV multispectral dataset. Accurate a prerequisite site-specific management to enable sustainable agriculture. This work aims analyse (spectrally) in wheat fields integrating Unmanned Aerial Vehicle (UAV), imagery machine learning techniques. 18 widely-used Spectral Indices (SIs) generated from 5 raw bands. Then various feature selection algorithms adopted improve model simplicity empirical interpretability. Forest classifier Bayesian hyperparameter optimization preferred as classification algorithm. Image also incorporated into map Guided Filter. The developed framework illustrated experimentation case naturally infected field Nottinghamshire, United Kingdom, where images were captured RedEdge on-board DJI S-1000 at altitude 20 m ground resolution 1.16 cm/pixel. Experimental results show that: (i) good result (an average precision, recall 93.8%, 93.0%) achieved system; (ii) most discriminating SI triangular greenness (TGI) Green-NIR, while wrapper can not only reduce number but achieve better than using all 23 features; (iii) filter helps performance noises.