A rapidly deployable approach for automated visual weed classification without prior species knowledge

This thesis presents the first work in automated, rapidly deployable species-wise classification of weeds for agricultural robotics where the species are unknown prior to arrival at the field. It focuses on extracting meaningful visual features to describe plants, unsupervised clustering to find visually similar plants, and rapid labelling to identify plant species and train classifiers with low manual effort. The approach enables agricultural robots to provide species-specific treatment more readily on a greater variety of fields