A Method for Identifying Easy-to-Pinch Axillary Buds

Robotics is essential for organic-farming expansion in Japan because of the climate and the declining and aging agricultural population. The demand for routine farming-task automation is growing in small-scale, open-field farms; these account for a large proportion of domestic farmland. Bud-pinching is one such routine task that involves the removal of axillary buds that are unnecessary for crop growth. If bud-pinching is not performed at an appropriate time, crop development may be affected negatively. Variability in the timing of axillary-bud emergence necessitates daily management. The detection and identification of typical stems has been addressed by conventional research; however, that of small and irregularly shaped axillary buds has not been studied sufficiently, and pinchable axillary-bud selection from densely clustered buds has not yet been automated. Plant stems were stably detected even under wind-induced movement, using a support vector machine (SVM) that focused on branch points, in our previous study. Branch points were identified as main stems, lateral branches, or axillary buds, and model effectiveness was experimentally evaluated and demonstrated. Axillary-bud detection has limitations caused by their small size, variable orientation, and tendency to be obscured by the surrounding flower buds, petals, or leaves. This study aimed to improve prediction accuracy by detecting and identifying the easy-to-pinch axillary buds. Our proposed method selected image regions by designing generic-template—containing branch points with sufficient space (for accessibility by the robot-arm end-effector)—and identified the stems within only those regions using our SVM model. The effectiveness of the proposed method was experimentally evaluated and demonstrated.