基于图像处理的秧苗均匀度合格率检测  

Seedling uniformity detection based on image processing

作　　者：马立新 夏利利 刘璎瑛[2] 李芃萱 朱伟 MA Li-xin;XIA Li-li;LIU Ying-ying;LI Peng-xuan;ZHU Wei(Jiangsu Agricultural Machinery Testing Station,Nanjing 210017,China;College of Artificial Intelligence,Nanjing Agricultural University,Nanjing 210031,China;College of Engineering,Nanjing Agricultural University,Nanjing 210031,China)

机构地区：[1]江苏省农业机械试验鉴定站,江苏南京210017 [2]南京农业大学人工智能学院,江苏南京210031 [3]南京农业大学工学院,江苏南京210031

出　　处：《江苏农业学报》2022年第2期387-393,共7页Jiangsu Journal of Agricultural Sciences

基　　金：江苏省现代农机装备与技术示范推广项目(NJ2019-25)。

摘　　要：插秧机试验鉴定时要人工测定插秧前秧苗培育均匀度,为提高鉴定效率,本研究提出基于图像分割和形态学操作的秧苗均匀度合格率自动检测方法。首先将获取的秧苗根茎部图像在2G-R-B颜色空间进行灰度化处理,阈值分割后进行形态学操作,完成面积阈值和形状阈值的二次分割,得到只含有水稻秧苗的二值化图像;其次根据移距和秧苗深度确定取样方格大小,按方格大小进行图像划分,选取图像中间部分的20个小格,输出每小格内的秧苗数量,与农艺要求进行比对,符合要求记作该小格合格;最后根据DG/T 008-2019《农业机械推广鉴定大纲水稻插秧机》,计算3个不同苗盘图像的合格方格数,得到插秧前秧苗均匀度的合格率。结果表明,采用图像处理方法可以实现插秧前秧苗的均匀度合格率计算,秧苗统计的准确率可以达到97.95%,方格检测的准确度可以达到96.67%。处理每幅图像的平均耗时为2.461 s,大大提高了检测效率。The cultivation uniformity of seedlings before transplanting needs to be measured manually in the identification process of seedling transplanter.To improve the identification efficiency,an automatic detection method for the pass rate of seedling uniformity was proposed based on image segmentation and morphological operation.Firstly,images of the seedling roots and stems were processed using graying method in 2G-R-B color space,and were processed by morphological operations after threshold segmentation by Otsu method.Binary images containing rice seedlings only were obtained after secondary division of the area threshold and shape threshold.Secondly,sizes of the sampling squares were determined by shift distance and seedling depth,then the images were divided into squares by the size.After selecting 20 small grids in the middle part of the images,the number of seedlings in each small grid was output and compared with agricultural requirements,and the confirmed small grid was recorded as qualified.Lastly,based on DG/T 008-2019<Outlines for agricultural machinery popularization identification:rice transplanter>,number of the qualified squares from images of three different seedling plates were counted,and the pass rate of seedling uniformity before transplanting was calculated.The results showed that,the pass rate of seedling uniformity could be calculated before transplanting by image processing methods,the statistic accuracy of seedlings could reach 97.95%,and the accuracy of the grid detection could reach 96.67%.The average time for processing each image took 2.461 s,which improved the detection efficiency greatly.

关 键 词：秧苗茎部图像 图像分割 形态学操作 均匀度检测 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] S511.01[自动化与计算机技术—控制科学与工程]