基于机器视觉的蔬菜种子分拣系统

发布时间：2018-12-09 08:00

【摘要】：本文以种子分拣系统为研究背景,以大豆种子为研究对象,采用VC6.0、Opencv以及MATLAB等为研究工具,从大豆种子图像的预处理过程、边缘提取过程、粘连豆类种子的分割、形状特征分量、颜色特征分量、机械手驱动运行以及PC机与PLC之间的通讯过程进行了深入的研究,对机器视觉筛选种子的研究具有一定的实际意义。研究取得了以下结果：1.大豆种子在光源下采集得到,在普通白炽灯下拍摄能够获得种子图像的大部分特征。试验中选用的背景色为白色,这样由于光源是垂直照射而导致种子中间部分反光强烈,用最大类间算法提取目标物时,容易将中间部分和背景色混为一体。因此,对图像要进行一些预处理。本文用两种方式进行研究,一种是进行灰度值的压缩,另一种是提取B、G、R三通道的二值化,为了后期颜色特征的分析,这里选用了三通道二值化法。2.为了识别干瘪种子,分析对比了Roberts算子、Sobel算子、Prewitt算子和Canny算子边缘提取的几种算法,选择合适的方法对种子进行边缘提取,可以达到筛选出干瘪种子的目的。在筛选过程中不可避免的遇到两粒或者是多粒种子之间相互粘连而造成误判断的现象。通过对粘连豆粒种子图像采用先腐蚀方式使得粘连的豆粒种子分割开来,然后对腐蚀后的单体豆类种子进行相同次数的膨胀处理,这样得到的二值化图像基本上保持了原图像的形状特征,对以后提取诸如面积、周长以及圆形度等其他一些参数基本上没有影响,用此算法解决了豆粒的粘连分割问题。3.定义并提取了18种大豆种子的特征分量,不仅在形状方面做了比较,而FL在颜色方面也作了相应的比较,提高了识别率。对标准的特征量用数学处理软件Minitab进行分析,确定其是否符合正态分布,对于符合正态分布的按照3σ区间进行定义合格种子,不符合正态分布的提出±20%的偏差,这样基本上能筛选出不合格的豆类种子。最后应用VC编程软件完成了种子分拣界面的编写,实现了对不合格种子的标识。4.对种子输送装置用变频器实现变频调速的控制；对机械手驱动装置进行分析,确定用PLC高速脉冲的方式驱动伺服电机,并在理论上模拟了机械手对种子三个位置的抓取动作；研究了PC机和PLC的通讯过程,能实现PC机和PLC最基本的通讯过程,理论上验证了机械手分拣大豆种子的可行性。综上所述,本课题的研究为种子筛选实现自动化、科学化提供了一系列的理论基础和技术支持,克服了人工检测费事费力的缺点,对种子实现高速、准确以及无损检测具有一定的实际应用价值。
[Abstract]:In this paper, the seed sorting system was used as the research background, the soybean seed was used as the research object, VC6.0,Opencv and MATLAB were used as the research tools, from the pretreatment process of the soybean seed image, the edge extraction process, the segmentation of the adhesion bean seed. The shape feature component, color feature component, manipulator driving operation and the communication process between PC machine and PLC are deeply studied, which is of practical significance for machine vision screening of seeds. The results are as follows: 1. Soybean seeds were collected under light source and captured under ordinary incandescent lamp to obtain most of the characteristics of seed images. The background color selected in the experiment is white, so the reflection of the middle part of the seed is strong due to the vertical illumination of the light source. It is easy to mix the middle part and the background color together when extracting the object with the maximum class algorithm. Therefore, the image should be preprocessed. In this paper, two methods are used to study, one is to compress the gray value, the other is to extract the binarization of the three channels of BGN R. In order to analyze the color characteristics of the later stage, the method of tri-channel binarization is chosen. 2. In order to identify dried seeds, several algorithms for edge extraction of Roberts operator, Sobel operator, Prewitt operator and Canny operator were analyzed and compared. In the process of screening, it is inevitable to encounter the phenomenon of two or more seeds conglutinating each other, resulting in misjudgment. By using the method of corrosion to separate the conglutinated soybean seeds, the monospecific seeds were treated with the same number of expansions. The binary image obtained by this method basically maintains the shape feature of the original image, and has no effect on the extraction of other parameters such as area, perimeter and roundness in the future. The algorithm is used to solve the adhesion segmentation problem of soybean grains. 3. The characteristic components of 18 kinds of soybean seeds were defined and extracted. Not only the shape but also the color of FL were compared, which improved the recognition rate. The standard characteristic quantity is analyzed by mathematical processing software Minitab to determine whether it conforms to the normal distribution, and the deviation of 卤20% for the definition of qualified seed according to the 3 蟽 interval is not in accordance with the normal distribution. This basically can screen out the unqualified bean seed. Finally, VC programming software was used to complete the preparation of seed sorting interface, and the identification of unqualified seeds was realized. 4. 4. The frequency conversion speed control of seed conveying device is realized by frequency converter, the driving device of manipulator is analyzed, the servo motor is driven by PLC high speed pulse, and the grasping action of manipulator to three positions of seed is simulated theoretically. The communication process between PC machine and PLC was studied. The basic communication process between PC machine and PLC was realized, and the feasibility of sorting soybean seeds by manipulator was verified theoretically. To sum up, the research of this subject provides a series of theoretical foundation and technical support for the realization of automation and scientific seed screening, overcomes the disadvantages of manual detection, and achieves high speed for seeds. Accurate and nondestructive testing has certain practical application value.
【学位授予单位】：天津科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S63;TP391.41

【相似文献】