Annotated cotton field image dataset for contamination detection and removal

Contamination found in cotton fields comprises plastic and other trash objects that can mix with the cotton when harvested. The values of the cotton bales are greatly diminished if harvesting machinery incorporates such items into the bales of ginned lint. This dataset was created to enable the development of vision systems capable of detecting contamination prior to or during harvest. The dataset is comprised of arial images of cotton fields having four classes of contamination elements annotated using locating boxes. The four classes include bags, bottles, cans, and trash. In addition to plastic bags, the bag class also includes any plastic sheet-like materials. Bags and sheets can become intertwined with the cotton plants. However, both the bottle and can elements typically stay on the ground level in the field and are less likely to become intertwined with the plants than the bag elements. In this dataset all the bottle and can elements are drink containers. The trash class includes any foreign materials not included in the other three classes. The dataset includes three growing seasons of cotton fields in Mississippi in 2021, 2022, and 2023. All the contamination was placed in the field just prior to imaging and removed immediately after. The dataset includes square image tiles of size 720 x 720 pixels with each image having an associated extensible markup language (XML) file containing the annotation data. Vision systems trained using this data can be a first step toward addressing contamination and enhancing the value of the crop. The dataset can also serve as a ground truth to test vision systems.