摘要
在进行红豆副产品深加工时,首要任务是根据外观品质粗略评估红豆原材料是否优良。传统的红豆品质和缺陷检测方法存在人工识别主观性强与分拣困难的问题。本文通过手机拍摄红豆数字图片,基于YOLOV4-Tiny深度学习网络和迁移学习思想自动提取图像特征,进行红豆外观品质识别研究,最终对于优质红豆识别准确率可达86.28%,劣质红豆识别准确率可达89.00%,模型mAP值可达90.78%。同时测试模型速度,对于红豆视频流检测可达46 fps,符合实际应用场景,为红豆外观品质识别与分级识别提供了新方案。
In the deep processing of red bean by-products,the primary task is to roughly evaluate the quality of raw materia.The traditional red bean quality and defect detection has the problems of strong subjectivity of manual identification and difficulty in sorting.In this study,the mobile phone was used to take digital images for red beans,and the image features were extracted based on the YOLOV4-Tiny network and transfer learning to identify the appearance quality.The results showed that the recognition accuracy of highquality red beans could reach 86.28%,and that of poor-quality red beans could reach 89.00%,and the model mAP value reached90.78%.At the same time,the speed of the model was tested as well,the detection of red bean video stream could reach 46 fps,which conformed to the actual aplication scenario.This paper provided a new scheme for the recognition of the appearance quality and classification of red bean.
作者
康烨
邱金凯
佟尚谕
许秀英
KANG Ye;QIU Jinkai;TONG Shangyu;XU Xiuying(College of Engineering,Heilongjiang Bayi Agricultural University,Daqing 163319,China;College of Information and Electrical Engineering,Heilongjiang Bayi Agricultural University,Daqing 163319,China)
出处
《内蒙古农业大学学报（自然科学版）》
CAS
2022年第4期82-87,共6页
Journal of Inner Mongolia Agricultural University(Natural Science Edition)
基金
黑龙江八一农垦大学校内培育课题项目(XZR2017-10)
