摘要
为了提高大豆千粒重的测定效率和精度,提出一种基于机器视觉的有效分割多层重叠颗粒和记数方法。将由摄像机采集的大豆颗粒图像经预处理后,提取所有的颗粒块区域;寻找颗粒块区域的拓扑形状特征欧拉数和形状特征向量:颗粒块轮廓边缘上凹点、颗粒块类圆近似核心;采用支持向量机分类法智能识别颗粒叠加类型,将其分为串重叠、多个并粘连和两层并重叠,最后实现自动分割。结果表明:该方法能有效解决两层重叠或深度粘连类圆大豆颗粒的准确计数。
In order to improve the efficiency and precision for measuring the mass of 1 000 soybean grains, a novel new method which can effectively segment overlapping granule and count it based on the machine vision technique was proposed. Firstly collecting the products grain image and be preprocessed,then distilling the region of all overlapping grain;The concave points, grains＇ approximate center of area and the euler number of topology shape were found out as the feature vectors of the overlapping area;the types of overlapping particle intelleetively were finally identified into serial, parallel and double-deck overlapping types by the classification of Support Vector Machine, and automatically divided overlapping granule into lots of single grains. The experimental results show that the paper can effectively count the double-deck overlapping or profoundly cohesive soybean grains.
出处
《大豆科学》
CAS
CSCD
北大核心
2009年第1期151-155,159,共6页
Soybean Science
基金
江苏大学江苏省现代农业装备与技术重点实验室开放基金资助项目(NZ200709)
关键词
支持向量机
分割
计数
大豆
Support Vector Machine
Segment
Count
Soybean
