基于骨架提取算法的作物表型参数提取方法

作物育种表型分析研究中,株型参数的获取多以人工测量为主,比较耗时费力。该文基于最小二乘法和遗传算法相结合,提出了一种用于计算作物表型参数的骨架提取方法。以玉米作物为例,首先为去噪后的作物二值图像进行单像素细化,利用角点检测归类算法,检测出特征点;依据骨架图像茎叶角点,利用图像分割将作物茎和叶分离,并对应图像中作物的茎和叶骨架,得到玉米作物空间离散点的实际三维坐标;融合最小二乘法和遗传算法,绘制出离散点的空间拟合曲线,即茎和叶的平滑骨架,从而提取出玉米作物的表型参数。田间试验分析表明,使用该算法能够有效地得到玉米作物的平滑骨架,而且与前人方法相比,测量得到表型参数中,株高误差减小了35%,叶长误差减小了70%,叶倾角误差减小了20%,有效地提高了作物表型参数的测量精度。该研究为提高作物表型参数尤其是株型参数精度提供了参考。

Crop phenotypic parameters extraction method based on skeleton extraction algorithm

In view of the problem that obtaining maize plant type parameters mainly depends on the artificial measurement at present in China, which is time-consuming and strenuous, a method to extract phenotypic parameters based on range images was proposed.Based on the corn crop, the algorithm first carries on the single pixel refinement and detection of feature points after denoising the binary image of crops; then stalk and leaves are separated according to the characteristics of the crops corresponding the space coordinates of pixels; according to the least square thought, the objective function and fitness function of genetic algorithm are improved to get the space curves which fit the space discrete points, finally the crop phenotypic parameters could be extracted through the space curves which are also the smooth skeleton of the stem and leaves.Compared with literature 9, field experimental results of the improved algorithm showed that the error of plant height is reduced by about 35%, the leaf length error reduced by 70% and the leaf angle error reduced by 20%.Experimental results demonstrate that the algorithm effectively improves the phenotypic parameters measurement precision, and provides the technical support for 3D model reconstruction of crops.