摘要
为建立食用菌图像识别模型,该研究以五种常见的食用菌菇(白玉菇、海鲜菇、蟹味菇、白蘑菇、香菇)为对象,利用图像采集系统分别对五种菌菇采集70幅图像,经图像处理后提取5个形态特征(面积、周长、矩形度、宽长比、圆形度)、6个颜色特征(’R、’G、’B、H、S、V)和4个纹理特征(角二阶矩、能量、对比度、熵),共计15个特征数据。采用支持向量机(Support vector machines,SVM)、随机森林(Random forest,RF)和卷积神经网络(Convolutional neural networks,CNN)构建识别模型,以全特征、形态特征、颜色特征、纹理特征分别对五种菌菇进行建模和验证。结果表明:RF模型在建模和预测精度上高于SVM模型,且以全部特征构建模型识别结果最佳,可作为构建食用菌图像数据库的识别模型。
In order to establish image recognition modle of edible fungi,five kinds of edible fungi((Lentinus edodes,Hypsizygus marmoreus,Agaricus bisporus,Hypsizigus marmoreus,White Hypsizygus marmoreus)were selected as research objects in this study.Computer vision system was constructed to acquire 350 images of five kinds of edible fungi.After image processing,morphological characteristics(area,perimeter,rectangularity,width to length ratio,roundness),color in the space of RGB(red,green,blue)and HSV(hue,saturation,value),texture fearture(angular second moment,energy,contrast,entropy)were extracted with a total of 15 parameters acquired.SVM(Support vector machines),RF(Random forest)and CNN(Convolutional neural networks)were used to build the discriminating models for eatable fungi type.Modeling and verification were carried out for five species of eatable fungi with full features,morphological features,color features and texture features respectively.The results showed that,the all features can well reflect edible fungi features.RF models are able to identify and distinguish eatable fungi with satisfactory results,which can provide technical support for practical application.
作者
王振杰
张乐乐
李双芳
章萩月
付倩倩
刘生杰
WANG Zhengjie;ZHANG Lele;LI Shuangfang;ZHANG Qiuyue;FU Qianqian;LIU Shengjie(College of Information Engineering,Fuyang Normal University,Fuyang Anhui 236041,China;College of Biology and Food Engineering,Fuyang Normal University,Fuyang Anhui 236037,China)
出处
《阜阳师范大学学报（自然科学版）》
2021年第4期42-48,共7页
Journal of Fuyang Normal University:Natural Science
基金
安徽省教育厅自然科学重点项目(KJ2019A0951)
阜阳师范大学信息工程学院校级项目(2019FXGZK03)。
