Optimizing UAV seaweed mapping through algorithm comparison across RGB, multispectral, and combined datasets

Abstract

The use of unmanned aerial vehicles (UAVs) with off-the-shelf RGB and multispectral sensors has expanded for environmental monitoring. While multispectral data enables analysis impossible with RGB, visible range cameras have benefits for large-scale habitat mapping. This research compared RGB, multispectral, and fused RGB-multispectral data from UAVs for seaweed mapping along the Irish coast. Three classification algorithms – Random Forest, Maximum Likelihood Classifier and Support Vector Machines – were tested on the three datasets to compare accuracies for seaweed species delineation and percent cover estimation. The RGB sensor effectively classified broad intertidal classes, but struggled differentiating some seaweed species. Multispectral data significantly improved species-level classification accuracy but tended to overestimate the presence of red and green algae. Fusing the RGB and multispectral data improved species classification accuracy over multispectral and RGB images. The results demonstrate the benefits of RGB sensors for broad habitat mapping and cover estimation, and multispectral for detailed species delineation. Fusion of the two sensor types enhances the strengths of both. This highlights the potential for UAVs paired with off-the-shelf visible range and multispectral cameras to provide detailed, accurate, and affordable change monitoring of intertidal seaweed habitats.