Augmented-Weed Detection dataset (modified-YOLOv7 )

The original images were captured at a resolution of 4096 × 3072 pixels and were subsequently tiled into smaller image patches to improve the detection of small objects, particularly weed instances. All images were annotated in YOLO format with two classes: weed and cotton.

To address class imbalance, since weed instances were less represented than cotton, selected images containing weed objects were augmented while the original images were retained. The augmentation process included horizontal flipping and brightness adjustment, which helped increase the representation of minority-class samples and improve dataset balance for model training under varying field conditions.