基于机器视觉的马铃薯智能分选方法与实现  

Implementation of intelligent potato grading method based on computer vision

作　　者：邓立苗[1] 杜宏伟[2] 徐艳[1] 韩仲志[1] 

机构地区：[1]青岛农业大学理学与信息科学学院,山东青岛266109 [2]青岛农业大学机电工程学院,山东青岛266109

出　　处：《中国农机化学报》2015年第5期145-150,共6页Journal of Chinese Agricultural Mechanization

基　　金：国家自然科学基金(31201133);青岛市科技发展计划(11-2-3-20-nsh)

摘　　要：为实现马铃薯智能检测与自动分级,提高马铃薯分级效率,本文在现有水果机械分选机的基础上,加装机器视觉系统和智能分级控制系统,提出马铃薯外观品质检测算法,实现马铃薯智能分选系统。首先下位机发送信号给上位机机器视觉系统控制摄像头拍照;然后上位机根据马铃薯形状、颜色和缺陷特点,采用近似椭圆法进行形状检测,采用逐点检测法检测绿皮区域,采用自适应阈值分割法分离缺陷区域,并以缺陷面积比进行缺陷检测;最后上位机将检测结果通过串口发送给下位机,分级执行器执行分级结果将次品拣出,再配合机械分选的压力传感器信号进一步实现正常品的重量分级。经测试:本文提出的分级检测算法对形状、绿皮和缺陷的检测正确率分别为93.3%、94.1%和88.3%,综合检测准确率可达到90%。本文构建的分级系统运行稳定,每秒可分选25个马铃薯,基本满足马铃薯实时分选的需求。In order to realize real-time potato inspection and improve the efficiency of potato grading, the implementation of the intelligent potatoes grading system based on computer vision was designed which was constructed by adding visual inspection system and grading control system on the existing fruit mechanical grading machine. In the meanwhile potato detection algorithm and grading control method were presented. First, the lower computer controls the camera to take photos by sending signal regularly to the visual inspection system on the upper computer. And then the upper computer will detect the potato from three aspects： shape, color and defect. The shape of potato is detected by ellipsoid method, which can depict the degree of potato that approaching the ellipse. Green skin is detected by R and H component value and realized by point-by-point detection method. The defect area is separated by adaptive threshold segmentation and defect area ratio is defined to measure whether there is defect on the potato surface. At last detection result will be sent to the lower computer by serial port and then control the grading actuators to pick up the inferior-quality products. 470 potatoes were selected randomly as testing samples to verify the accuracy and speed of the detection algorithm. The accuracy rates of the detection of shape, green and defect were 93.3%, 94.1% and 88.3%, and the comprehensive accuracy rate of all features reached 90%. This system can grade 25 potatoes per see ond and run steadily, which can meet the needs of the real-time detection. It has a positive significance for constructing an automatic detection and grading system of potato.

关 键 词：机器视觉 检测 椭圆度 自适应阈值 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]