基于机器视觉和YOLOv5的油菜角果表型参数分析  

Analysis of phenotypic parameters of rapeseed silique based on machine vision and YOLOv5

作　　者：陈威 朱怡航 顾清[2] 林宝刚[3] 张小斌[2] CHEN Wei;ZHU Yihang;GU Qing;LIN Baogang;ZHANG Xiaobin(College of Mathematics and Computer Science,Zhejiang A&F University,Hangzhou 311300,China;Institute of Digital Agriculture,Zhejiang Academy of Agricultural Sciences,Hangzhou 310021,China;Institute of Crop and Nuclear Technology Utilization,Zhejiang Academy of Agricultural Sciences,Hangzhou 310021,China)

机构地区：[1]浙江农林大学数学与计算机科学学院,浙江杭州311300 [2]浙江省农业科学院数字农业研究所,浙江杭州310021 [3]浙江省农业科学院作物与核技术利用研究所,浙江杭州310021

出　　处：《浙江农业学报》2024年第6期1379-1388,共10页Acta Agriculturae Zhejiangensis

基　　金：浙江省农业旱粮新品种选育重大科技专项“油菜新品种选育”(2021C02064-2-3);浙江省农业科学院生物育种融通计划“作物表型高通量获取与智能应用平台建设”(2022SWYZ)。

摘　　要：为了更加高效和准确地获取油菜角果表型参数,在图像处理技术和深度学习算法的基础上,以迎春一号油菜角果为实验材料,综合考虑油菜育种对角果外观表型参数的需求,提出了一种基于机器视觉的油菜角果表型分析方法:利用图像处理技术实现了油菜角果的柄喙长度、果身长度、果身宽度、弦长、弧长、面积等外观表型性状的提取,使用YOLOv5对单角果籽粒进行无损计数。对角果实物及标定物进行测量验证,结果表明,图像分析出的角果表型指标与人工实际测量值无显著性差异(P>0.05),决定系数(R^(2))均大于0.96,均方根误差(root mean square error,RMSE)均小于3 mm,平均绝对值误差(mean absolute error,MAE)均小于2.80 mm,平均绝对百分比误差(mean absolute percentage error,MAPE)均不超过4%。标定物直径最大RMSE为0.3 mm,MAE均小于0.28 mm,MAPE均小于2.00%,面积指标最大RMSE为12.09 mm^(2),MAE均小于11.56 mm^(2),MAPE均小于5%。YOLOv5识别出的籽粒数与实际值无显著性差异(P>0.05),R^(2)为0.987,RMSE为0.68粒,MAE为0.27粒,MAPE为1%。该研究的油菜角果表型分析方法操作简单、成本较低,能有效地减少人工测量的误差,提高获取表型信息的可靠性和油菜育种工作的效率,为油菜表型信息的定量化分析提供了一定的参考。To obtain the phenotypes of rapeseed siliques more efficiently and accurately,a machine vision-based method for analyzing the phenotypes of rapeseed siliques was proposed using image processing technology and deep learning algorithms.Siliques of rapeseed cultivar Yingchun No.1 were used as materials.The need to acquire morphological phenotypes of siliques is considered for rapeseed breeding.Image processing technology was used to extract the morphological phenotypes of the rapeseed siliques,including pedicel length,silique length,silique width,chord length,arc length,and area.YOLOv5 was used to perform nondestructive silique grain counting.Measurements and verifications were made on the siliques and calibration objects.There was no significant difference(P>0.05)between the phenotypic indexes of siliques assessed by image analysis and the actual measured values.The R^(2)was more than 0.96,the root mean square error(RMSE)was less than 3 mm,the mean absolute error(MAE)was less than 2.80 mm,and the mean absolute percentage error(MAPE)was less than 4%.The calibration object diameter had a maximum RMSE of 0.3 mm,the MAE was less than 0.28 mm,and the MAPE was less than 2.00%.The area index had a maximum RMSE of 12.09 mm^(2),the MAE was less than 11.56 mm^(2),and the MAPE was less than 5%.There was no significant difference between the number of grains identified by YOLOv5 and the actual value(P>0.05),R^(2)was 0.99,RMSE was 0.68,MAE was 0.27,and MAPE was 1%.The method for analyzing the phenotypes of rapeseed siliques proposed in this study is easy to operate and labor-saving.It can effectively reduce the manual measurement error,improve the reliability of obtaining phenotypic information,and increase the efficiency of rapeseed breeding work.It also provides particular guidance for the quantitative analysis of rapeseed phenotypic information.

关 键 词：油菜角果 表型 图像处理 籽粒计数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]