Cotton Weed Database Development and Deep Learning Model Benchmarking (2021, Cotton Inc., Cary, NC)

This research is to develop a large-scale, public image database of cotton weeds in the field and benchmark deep learning models for weed detection, with the ultimate goal to enhance computer vision-based robotic weeding for cotton. Three sub-objectives are to: 1) Acquire color images in the weed-infested cotton fields under natural light conditions at varied crop growth stages; 2) Expand the image database algorithmically and annotate images for the weed instances at image and pixel levels; 3) Provide benchmarking results of weed classification and detection for the image dataset using deep learning algorithms.