Utilizing Machine Vision for Real‐Time Robotic Arm‐Assisted Singulation of Wash‐Root Nursery Seedlings

ABSTRACT

Manual transplanting of wash‐root seedlings is a common practice among Indian farmers, but it involves significant labor and drudgery. To automate the transplanting process, this study presents a novel system that detects and centers wash‐root nursery seedlings for robotic arm‐assisted singulation, simulating human actions. Using pretrained architectures and data sets with 200 and 1000 images, a detection model was created. Uniquely, the study emphasizes how important it is to expand the data set to enhance model performance. Without using data augmentation approaches, the model was trained to detect and center seedlings by gathering a larger data set of 1000 photos. During laboratory evaluations, the larger data set significantly enhanced the model's detection accuracy and generalization ability, enabling precise centering of seedlings for robotic singulation. Performance was evaluated at three levels of epochs and data set ratios, with the model trained on 1000 images achieving an Average Precision of 63.4%, a High Confidence Score of 89%, and a Validation Loss of 3.19. Real‐time evaluations using the selected model demonstrated exceptional centering capability, with a perfect visual rating of 100% for its ability to accurately center and singulate seedlings. This study offers a viable approach for automated transplanting in agriculture by highlighting the efficiency of using pretrained models, transfer learning, and cautious data set extension to improve robotic seedling handling.