Estimation of Tiller Number in Rice Using a Field Robot and Deep Learning

Tiller number, an important growth parameter for rice cultivation, is still being assessed manually. This work investigated the influence of dataset composition on performance deep learning models tiller number estimation in rice. Four datasets were constructed early tillering, active and maximum tillering by applying concepts mixed varieties, class balance, data augmentation. YOLOv4 trained to estimate numbers using each dataset. Then their was evaluated. Results demonstrated that with created a combination variety, augmentation showed best estimating at three stages mAP range 68.8–86.4 %.