In view of the inaccurate extraction of feature information for plant 3D point cloud simplification, a judging method of the region complexity was proposed to optimize the curvature simplified algorithm by combining the error of the local curvature and the normal vector angle. Comparing the standard deviation of curvature and the normal vector angle with the threshold of K neighborhood in each data point, the complexity of the local area is determined. Then, different reduction rates are used to determine whether the neighborhood points are retained or not, and their retention probabilities are counted. Finally, the trade–off between data points is determined through the overall reduction rate and retention probability. Compared with the traditional simplified algorithm applied to the plant point cloud, the results show that the leaf vein characteristics of the leaf are more obvious, and the deviation is reduced by more than 25% under the similar reduction rate.