摘要
针对马铃薯表面缺陷检测快速准确的需求,提出一种基于机器视觉与YOLO算法的马铃薯表面缺陷检测方法。应用这一方法,构建马铃薯表面缺陷图片数据集,对原始数据集进行图像增广;通过二分K均值聚类算法进行目标框聚类分析,采用分步训练方式优化学习权重。试验结果表明,所提出的基于机器视觉与YOLO算法的马铃薯表面缺陷检测方法可以有效实现马铃薯表面缺陷的快速、准确检测,平均识别精度达到99.46%,对腐烂、发芽、机械损失、虫眼、病斑检测的精度均高于98%,单幅图片识别时间约为29 ms。
Aiming at the requirement of fast and accurate detection of potato surface defect,a potato surface defect detection method based on machine vision and YOLO algorithm was proposed.This method is applied to construct an image data set of potato surface defects,and perform image augmentation on the original data set.A dichotomous K means clustering algorithm is used to perform target box clustering analysis,and a step by step training method is used to optimize learning weight.The test results show that the proposed potato surface defect detection method based on machine vision and YOLO algorithm can effectively realize the fast and accurate detection of potato surface defect,with an average identification accuracies of 99.46%,and the detection accuracies on rotting,sprouting,mechanical loss,insect eyes,and disease spots are higher than98%,and the recognition time of a single picture is about 29 ms.
出处
《机械制造》
2021年第8期82-87,共6页
Machinery
基金
国家自然科学基金资助项目(编号:51805280)
宁波市自然科学基金资助项目(编号:2019A610158)
宁波市科技创新2025重大专项(编号:2018B10005)。