基于机器视觉的荞麦剥壳性能参数在线检测方法  

On-line Measuring Method of Buckwheat Hulling Efficiency Parameters Based on Machine Vision

作　　者：吕少中 杜文亮[1] 陈震[1] 陈伟 苏日嘎拉图[1] LU Shaozhong;DU Wenliang;CHEN Zhen;CHEN Wei;Surigalatu(College of Mechanical and Electrical Engineering, Inner Mongolia Agricultural University, Huhhot 010018, China;College of Information Engineering, Inner Mongolia University of Technology, Huhhot 010080, China)

机构地区：[1]内蒙古农业大学机电工程学院,呼和浩特010018 [2]内蒙古工业大学信息工程学院,呼和浩特010080

出　　处：《农业机械学报》2019年第10期35-43,共9页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金项目(31260409);内蒙古自然科学基金项目(2014MS0310)

摘　　要：针对荞麦剥壳时不能随原料种类变化而适时调整砂盘间隙和转速的问题,提出一种基于机器视觉的荞麦剥壳性能参数在线检测方法,为荞麦剥壳机自适应最优控制提供数据反馈。采集快速滑落的荞麦剥出物图像,使用带二阶拉普拉斯修正项的边缘自适应插值算法对图像插值重建;对重建的浅蓝色背景荞麦剥出物图像N(B-R)灰度变换之后进行背景分割;生成距离骨架图像并对其邻域极大值滤波提取种子点,使用分水岭算法对种子点标记后的距离图像进行粘连分割;采用交互式方法标注已粘连分割的荞麦籽粒,然后使用已标注的荞麦籽粒训练BP神经网络。在线试验中,处理和识别一幅包含897个籽粒的1 824像素×1 368像素图像耗时4.79 s。未剥壳荞麦、整米和碎米的正确识别率分别为99.7%、97.2%和92.6%。结果表明,本文在线检测方法得到的出米率能够反映荞麦剥壳机组的剥壳性能,可为荞麦剥壳加工的自适应最优控制和智能化提供有效基础数据。In order to measure the efficiency parameters in the hulling process of buckwheat huller, an on-line measuring method based on machine vision to measure the efficiency parameters of buckwheat hulling was presented. The image of the fast sliding buckwheat grains was captured. N(B R) gray transformation was performed on the captured image of buckwheat grains with a light blue background, then with Otsu algorithm the background was segmented and a binary image of buckwheat grains was generated. A distance image of buckwheat grains was generated by performing Euclidean distance transformation on the binary image, a skeleton image of buckwheat grains was generated by performing thinning operation on that binary image, and then the corresponding pixel points of distance image and skeleton image were multiplied and a distance-skeleton image was generated. Seed points were extracted by performing neighborhood maximum filtering algorithm on the distance-skeleton image, the distance images were marked with seed points, and the touching buckwheat grains were segmented with watershed segmentation algorithm. An interactive labeling method was used to label the unshelled buckwheat, whole buckwheat rice, broken buckwheat rice and wrongly segmented buckwheat grains, and then the labeled buckwheat grains were used to train a BP neural network. In the online experiment, the recognition rates of unshelled buckwheat, whole buckwheat rice and broken buckwheat rice were 99.7%, 97.2% and 92.6% respectively and it took 4.79 s to process and recognize an 1 824 pixels×1 368 pixels image containing 897 seeds. The results showed that the rate of unbroken buckwheat rice can reflect the hulling efficiency of buckwheat huller and the running time met the need of on-line measurement.

关 键 词：荞麦 剥壳 在线检测 机器视觉 粘连分割 

分 类 号：S126[农业科学—农业基础科学] TP391.41[自动化与计算机技术—计算机应用技术]