A dataset of approximately 10,000 weed images was created for training by augmenting 541 raw images of three weed types—pigweed, purslane, and nutsedge—using the OpenCV library (Figure S3a). Roboflow software was used for image annotation and data preparation (Figure S3b). The dataset was split into 70% training, 20% validation, and 10% test data in VOC Pascal format prior to training. The image input size was set to 640 x 640 for YOLOX-s and 416 x 416 for YOLOX-nano.