基于计算机视觉的大米粒形识别方法

提出了利用计算机视觉系统代替人眼识别整粒大米和碎大米粒形的方法。设计了一套基于计算机视觉技术的大米粒形识别装置,采用灰度变换、阈值分割、平滑处理等图像处理方法获取大米的粒形图像,然后根据大米的粒形特点提取了米粒的面积、周长、长、宽等16个特征参数,采用主成分分析方法对提取的特征参数进行处理,以前三个主成分综合所有粒形特征参数,作为BP神经网络的输入特征值对网络进行训练和大米粒形识别。试验结果表明:该方法对整粒米识别的准确率为98.67%;对碎米识别的准确率为92.09%。

Recognition method of rice kernel shape based on computer vision

A method to recognise head rice and break rice based on computer vision system was proposed,which can replace human visual observation.A computer vision based detection system was developed,which uses image processing methods,such as gray transformation,threshold segmentation and smooth processing etc,to get the image of rice kernel shape.The kernel shape was characterized by sixteen parameters,such as the area,perimeter,length and width etc.The data from the kernel profile were processed by the principal components analysis method.The first three principal components,standing for all the parameters of the shape,were taken as the input eigenvalues to train the BP neural net.Recognition experiments show that the accurate ratio of distinguishing head rice is 98.67%,and the accurate ratio of distinguishing break rice is 92.09%.