Toward multi-class detection and localization of contamination in cotton fields: Comparing EfficientDet, Faster R-CNN ResNet50, SSD MobileNet V2, and YOLOv11

The economic value of a bale of ginned cotton decreases significantly if it contains plastic. One potential source of plastic is trash found in cotton fields. Detecting and removing plastic contaminating the fields will reduce the chance of plastic in the bale. Removal at the field is advantageous because harvest and postharvest machinery tend to break the materials into smaller pieces which are harder to remove. This paper compares multiple object detection architectures trained to find bags, bottles & cans, and general trash in a cotton field. Training and validation used a set of 15,200 images representing the 2021 and 2023 cotton crop in the Mississippi Delta with 80% of the images used for the training. Testing used 2526 images from the 2022 cotton crop. The best model checked on the validation set had a mean average precision of 0.91 and the best mean average recall was 0.75. Generalization to the 2022 crop test set did not perform as well with the best mean average precision falling to 0.83 and best mean recall falling to 0.57. This paper provides relevant information for reducing contaminants in cotton, minimizing the associated economic losses, and maintaining the United States cotton industry’s reputation.