基于机器视觉的大豆籽粒精选技术

为实现大豆精选模型的设计，选择东农405、东农410、东农634 共3个大豆品种，以正常豆、灰斑豆、霉变豆、虫蚀豆为研究对像，采用可脱离PC机独立工作的智能摄像头获取分析豆粒图像。通过动态阈值分割算法分离豆粒与背景，提取豆粒图像的形状、颜色、纹理3方面的特征参数15个。采用BP神经网络建立分类模型，模型平均识别准确率达98%。试验选择2000粒大豆样本对精选装置进行测试，测试结果显示：该装置对正常豆、灰斑豆、霉变豆和虫蚀豆的筛选精度分别达到98.3%、93.4%、92.2%、95.9%，筛选效率达到每分钟300粒，将机器视觉技术应用于大豆精选机的设计中是可行的。

Soybean seeds selection based on computer vision

To achieve the design of soybean selected model, the normal soybean, gray spot soybean, moldy soybean and worm-eaten soybean were chosen from the three kinds of soybean, (DongNong 405, DongNong 410, DongNong 634). The soybean images were obtained and analyzed with the intelligent camera working without PC. The 15 characteristic parameters of soybean image, such as shape, color and texture were extracted by means of separating the soybean and background with dynamic threshold separation algorithm. The average recognition accuracy of model reached 98% by building the BP neutral network classification model. The 2000 soybeans were used to test of selected device and the test results showed that the selected accuracy of normal soybean, gray spot soybean, moldy soybean and worm-eaten soybean were 98.3%, 93.4%, 92.2% and 95.9%, respectively. The selection rate was 300 per minute. Soybean selected device with machine vision technology was feasible.