Evaluation of hierarchical self-organising maps for weed mapping using UAS multispectral imagery

•Three hierarchical SOMs were tested on UAV images for S. marianum weed mapping.•Challenging discrimination between vegetation species with similar spectral reflectance.•Classifiers produced high quality maps of >98% agreement with the validation dataset.•The proposed neural networks provided the factors affecting S. marianum detection. Remote sensing has been used for species discrimination and for operational weed mapping. In the study presented here, the detection and mapping of Silybum marianum using a hierarchical self-organising map is reported. A multispectral camera (green-red-NIR) mounted on a fixed wing Unmanned Aircraft System (UAS) was used for the acquisition of high-resolution images of a pixel size of 0.1m, resampled to 0.5m. The Supervised Kohonen Network (SKN), Counter-propagation Artificial Neural Network (CP-ANN) and XY-Fusion network (XY-F) were used to identify the S. marianum among other vegetation in a field, with Avena sterilis L. being predominant. As input features to the classifiers, the three spectral bands of Red, Green, Near Infrared (NIR) and the texture layer were used. The S. marianum identification rates using SKN achieved an accuracy level of 98.64%, the CP-ANN achieved 98.87%, while XY-F was 98.64%. The results prove the feasibility of operational S. marianum mapping using hierarchical self-organising maps on multispectral UAS imagery.