摘要
针对香菇表型测量信息化程度低、人工测量费时费力的问题,提出了一种基于U-Net的香菇表型提取方法。通过采集香菇切面图像,建立数据集,实现了基于U-Net的香菇菌盖、菌柄、左右菌褶的分割,模型的平均交并比和平均像素准确率分别为85.00%和91.25%。结合质心法和最小外接矩形法实现了菌盖直径、菌盖厚度、菌柄长度、菌柄直径和菌褶宽度5个香菇表型参数的自动测量。与人工测量值相比,提出的方法在测量菌盖直径、菌盖厚度、菌柄长度、菌柄直径和菌褶宽度时,其平均绝对百分比误差分别为1.57%、5.01%、2.57%、5.47%、2.74%;均方根误差分别为0.12、0.08、0.09、0.10、0.06 cm;决定系数均大于0.90。结果表明,基于U-Net的香菇表型提取方法适用于香菇的表型测量,具备较高的分割准确率。
A U-net based method for extracting the phenotypes of shiitake mushrooms was proposed to solve the problems of the low level of informatization and the time-consuming and laborious manual measurement of phenotypes in shiitake mushroom.U-net based segmentation of caps,stipes,and left and right gills in shiitake mushroom was achieved by collecting the cross-sectional images of shiitake mushrooms and establishing a dataset.The results showed that the average intersection to union ratio and average pixel accuracy of the model was 85.00%and 91.25%,respectively.The automatic measurement of five parameters for phenotypes in shiitake mushrooms including the cap diameter,cap thickness,stipe length,stipe diameter,and gill width was achieved by combining the centroid method with the minimum bounding rectangle method.The method proposed had mean absolute percentage error(MAPE)of 1.57%,5.01%,2.57%,5.47%,and 2.74%in measuring the cap diameter,cap thickness,stem length,stem diameter,and gill width,with the root mean square error(RMSE)of 0.12 cm,0.08 cm,0.09 cm,0.10 cm,0.06 cm and all determination coefficients(R²)exceeding 0.90 compared with manual measurements.It is indicated that the U-net based method for extracting the phenotypes in shiitake mushrooms is suitable for measuring phenotypes in shiitake mushrooms with a high accuracy of segmentation.It will provide strong technical support for the sorting and grading of shiitake mushrooms.
作者
刘岩
刘欢
张恩帅
赵文瑞
祝梓涵
边银丙
梁秀英
LIU Yan;LIU Huan;ZHANG Enshuai;ZHAO Wenrui;ZHU Zihan;BIAN Yinbing;LIANG Xiuying(College of Engineering,Huazhong Agricultural University,Wuhan 430070,China;College of Plant Science and Technology,Huazhong Agricultural University,Wuhan 430070,China)
出处
《华中农业大学学报》
北大核心
2025年第4期302-309,共8页
Journal of Huazhong Agricultural University
基金
山东省重点研发计划项目(2022CXGC010609)
国家自然科学基金项目(U21A20205)。
关键词
香菇
图像分割
表型提取
U-net
shiitake mushrooms
image segmentation
extraction of phenotypes
U-net
