玉米籽粒激光切片定位技术

Clustering Method of Positioning for Maize Seed Laser-cutting Slices Combined with Spatial Constrains

摘要:

玉米籽粒形态各异、尺寸不一，精确定向和定位玉米籽粒的激光切片是实现高通量全自动玉米分子育种基因型分析的关键。应用机器视觉技术从玉米籽粒图像中准确识别玉米籽粒的特征区，以期实现上述操作。为描述像素所在空间的相关信息，设计一种相关面积占比滤波器。定义圆形掩模模板，根据单玉米籽粒的面积，确定模板尺寸。利用圆形模板筛选像素点数据，得到待分类数据集合。通过指定初始聚类中心，对数据执行二分均值聚类，得到尖端类和两个大端外凸角类的聚类中心。通过贴标签运算精选连通域，校正聚类中心的位置，生成尖端和大端外角特征区的精确标记。依据大端外凸角附近的两组插值点对，得到激光切割线的位置，利用尖端类定位点和玉米籽粒形心定位点确定玉米籽粒的夹持位姿。与SUSAN检测方法对比，表明了本文方法的有效性。

Abstract:

Maize seeds are of different shapes and sizes. It is the bottleneck that the maize seed lasercutting slices are orientated and positioned accurately for the molecular breeding genotype analysis to achieve high throughput with automation. The machine vision system means to recognize the maize feature regions for positioning lasercutting slices accurately in a single seed image. An area correlation filter was presented for describing the pixel with spatial constrain information. The definition of the round mask template was proposed for the seed morphological measurement. The round template size was determined by the area of a single maize seed. Some of pixel coordinate data were extracted to be classified from the target domains by the filtration of the area correlation filter. Through the bisectingmeans clustering with the specific initial clustering centers, the extracted data were divided into thin part class and thick part class. Also, their clustering centers were got, which were corresponded to the thin part class and two arc corner classes, respectively. The labeling partition operation was applied to the connected domains for finely adjusting and marking the centers of tip part and two arc corners. Finally, the coordinates of two interpolated pointpairs near the thick part were calculated. Through linking two pairs of interpolated points, the lasercutting lines were located with high precision. The seed clamping pose was determined according to the tip part center and the centroid of seed. Compared with SUSAN, SUSAN detector cannot be directly applied to locate the feature region of maize seed. The experimental results verified the effectiveness of the proposed method on yellow and white maize seeds.