高级检索
    黄本荣, 范兆烽, 王飞, 江逸昕, 马祥根, 肖光林, 詹德良, 吴善建, 黄嘉星, 温永仙. 基于VGG16-DenseNet集成模型的烤烟智能分级[J]. 中国烟草科学.
    引用本文: 黄本荣, 范兆烽, 王飞, 江逸昕, 马祥根, 肖光林, 詹德良, 吴善建, 黄嘉星, 温永仙. 基于VGG16-DenseNet集成模型的烤烟智能分级[J]. 中国烟草科学.
    HUANG Benrong, FAN Zhaofeng, WANG Fei, JIANG Yixin, MA Xianggen, XIAO Guanglin, ZHAN Deliang, WU Shanjian, HUANG Jiaxing, WEN Yongxian. Intelligent Classification of Flue-cured Tobacco Leaves based on Integrated with VGG16 and DenseNet[J]. CHINESE TOBACCO SCIENCE.
    Citation: HUANG Benrong, FAN Zhaofeng, WANG Fei, JIANG Yixin, MA Xianggen, XIAO Guanglin, ZHAN Deliang, WU Shanjian, HUANG Jiaxing, WEN Yongxian. Intelligent Classification of Flue-cured Tobacco Leaves based on Integrated with VGG16 and DenseNet[J]. CHINESE TOBACCO SCIENCE.

    基于VGG16-DenseNet集成模型的烤烟智能分级

    Intelligent Classification of Flue-cured Tobacco Leaves based on Integrated with VGG16 and DenseNet

    • 摘要: 为实现烤烟烟叶等级快速、准确的智能化识别,本研究基于手机拍摄的不同品种的烤烟烟叶正面、反面图像,构建了VGG16与DenseNet组合的新网络模型VGG16-Dense,并应用手机拍摄的翠碧1号、云烟87烤烟烟叶6个等级正反面图片,总共24类,验证该模型的有效性,同时与5个网络模型DenseNet121、ResNet50、AlexNet、VGG16和GoogLeNet进行比较。研究表明:VGG16-Dense网络模型在验证集的各评估指标(准确率、精确率、召回率、F1分数和平均损失值)均达到优秀值,在测试集的各评估指标较其他模型是最优的,准确率为92.71%,精确率为93.07%,召回率为92.71%, F1分数为92.72%,平均损失值为0.22,有较好的泛化能力, 错判较少。VGG16-Dense网络模型能够同时甄别品种、烤烟烟叶的正面和反面,这为烤烟烟叶智能分级应用于实际的初级烤烟收购提供了一个辅助帮助。

       

      Abstract: In order to achieve intelligent recognition of flue-cured tobacco leaves grade quickly and accurately, a new model (VGG16-Dense) that integrated with VGG16 and DenseNet was constructed based on the front and back picture of flue-cured tobacco leaves taken by mobile phone in this study. The validity of the model was verified by using twenty-four types of flue-cured tobacco leaves pictures to obtain by mobile phone for “Cuibi 1” and “Yunyan 87” . The model was also compared with five other network models: DenseNet121, ResNet50, AlexNet, VGG16 and GoogLeNet. The results shows that there were excellent values in all evaluation indicators (accuracy, precision, recall, F1-score and avg-loss) of validation set for VGG16-Dense. The evaluation indicators of test set for VGG16-Dense were optimal compared with that of other network models. The accuracy was 92.71%, the precision was 93.07%, the recall was 92.71%, the F1-score was 92.72%, and the avg-loss was 0.22. There was excellent generalization ability and fewer misjudgments for VGG16-Dense. VGG16-Dense network model can distinguish the front and back image, the species for flue-cured tobacco leaves at the same time.This provides an auxiliary help for the application of intelligent classification of flue-cured tobacco leaves in practice.

       

    /

    返回文章
    返回