基于视觉融合的大米外观质量检测系统研究

Research on Rice Appearance Quality Inspection System Based on Vision Fusion

在基于视觉处理的大米外观质量检测研究中,针对粘连米粒的分割以及垩白米粒的识别分别引起了研究者们的关注,但如何精准地进行这2种不同类型的分割并将分割结果进行处理和融合,进而提供一个完整的大米外观质量检测系统还鲜见报道。基于此,本研究首先提出了一个基于视觉融合的大米外观质量检测模型,该模型包括米粒轮廓线分割、米粒垩白区域分割、后处理3个模块。其次,在此模型的基础上,设计了一个自动检测系统,不仅能直接计算受检大米的垩白度、垩白粒率、碎米率、黄米率等指标,还能可视化检测结果。实验结果表明,该模型可以在像素级分割的基础上,实现大米外观质量的完整检测,对米粒垩白度、垩白粒率、碎米率和黄米率的平均识别准确率可分别达到94.5%、96.3%、97.9%和95.1%。

In the research of rice appearance quality detection based on visual processing,the segmentation of adhesive rice grains and the recognition and segmentation of chalky rice grains have attracted researchers attention in recent years.However,there is rare report on how to accurately perform these two different types of segmentation and fuse the results together to provide a complete rice appearance quality detection system yet.Based on this,in this paper,a rice appearance quality detection model based on vision fusion was proposed first,it included three modules:rice contour segmentation,rice chalkiness region segmentation and post-processing.Secondly,an automatic detection system based on this model was designed,it could not only calculated chalkiness degree,chalkiness grain rate,broken rice rate,and yellow rice rate directly,but also visualized detection results before and after visual fusion.The experimental results demonstrated that the proposed model could detect the appearance quality of rice completely on the basis of pixel-level segmentation,and the average recognition accuracy for chalkiness degree,chalkiness grain rate,broken rice rate,and yellow rice rate reached 94.5%,96.3%,97.9%and 95.1%,respectively.