Abstract: The flag leaf angle is one of key parameters for determining the rice yield, achieving accurate, efficient and in vivo measurement of flag leaf angle is significant to rice breeding, plant type research and production instruction. However, the stems in sample images are usually obscured, moreover, current algorithms cannot recognize flag leaf nodes and axes of diverging, bifurcate rice ears. Hence, a flag leaf node searching algorithm is presented, then the cluster center of rice ear and leaf is generated by a redefined clustering method in order to recognize the angles between rice ear and flag leaf. The leaf node searching algorithm quantifies the fuzzy localization of leaf node, and it is proved to be robust and accurate by experiment. The redefined K-means method is based on the statistical information of samples, it can solve the problem that current algorithms cannot measure angles between diverging, bifurcate rice ear and flag leaf. Furthermore, it is practical in measuring the intersection angles in various plants’ bifurcate form. Hence, the paper proposes a new thought of clustering in multi-axial data set. Experimental results show that, the algorithm had an error of 1.89% with low limitation, stronger robustness and higher degree of accuracy.

Key words: machine vision, flag leaf angle, in vivo measurement, clustering, node searching