摘要
目的:准确识别霉变玉米籽粒。方法:基于高光谱图像光谱变量和颜色特征建立霉变玉米籽粒识别的新方法。先对玉米籽粒图像进行图像分割和光谱变量、颜色特征提取,并根据颜色特征生成颜色直方图;将光谱变量和颜色直方图特征组成特征集合;通过距离函数对特征集合中所有特征的分析确定霉变玉米籽粒所属类别。结果:所提方法对霉变玉米籽粒类别的最大平均识别偏差为1.12,最佳平均识别准确率为97.59%;与基于高光谱图像+随机蛙跳+极限学习机的方法、基于高光谱图像+稀疏自动编码器+卷积神经网络的方法、基于高光谱图像+蚁群优化+BP神经网络的方法相比,研究所提方法对霉变玉米籽粒类别的识别准确率明显提高。结论:该方法可实现被测玉米籽粒样品是否霉变以及霉变程度的准确判断。
Objective:To identify mildew maize kernels accurately.Methods:A novel method to identify mildew maize kernels using spectral variables and color characteristics of hyperspectral images.Firstly,image segmentation,spectral variables and color features extraction were carried out on maize kernel images.Then,color features of maize kernel images were utilized to generate color histograms.Additionally,spectral variables and color histogram features were combined into a feature set.Finally,the distance functions were used to analyze the features in this feature set to identify mildew maize kernels.Results:For the proposed method,the maximum average identification deviation and accuracy for the mildew maize kernels were 1.12 and 97.59%,respectively.Compared with the method based on hyperspectral images+random frog+extreme learning machine,the method using hyperspectral images+colony optimization+BP neural network,and the method based on hyperspectral images+sparse auto-encoders+convolutional neural network,the identification accuracies of mildew maize kernels were significantly improved by the proposed method.Conclusion:The developed method can accurately identify whether the corn grain samples are mildew and the mildew degree of the maize kernel samples.
作者
李伟
赵雪晴
刘强
LI Wei;ZHAO Xue-qing;LIU Qiang(Huai'an Vocational Education and Teaching Research Office,Huai'an,Jiangsu 223001,China;Jiangsu University of Science and Technology,Zhenjiang,Jiangsu 212100,China;Suzhou University,Suzhou,Jiangsu 510632,China)
出处
《食品与机械》
北大核心
2022年第12期112-120,共9页
Food and Machinery
基金
国家科学基金面上项目(编号:5207729)
江苏自然科学基金项目(编号:21JS12903)。
关键词
玉米籽粒
霉变
识别方法
高光谱图像
光谱变量
颜色特征
maize kernels
mildew
identification method
hyperspectral image
spectral variables
color features
