基于不同卷积神经网络模型的陈皮、广陈皮图像的智能识别研究  

Research of Different Convolutional Neural Network Models for Intelligent Recognition of Chenpi and Guangchenpi

作　　者：陈志维[1,2] 唐珂轲 易智彪 许伟珊[2] 李知晓 CHEN Zhi-wei;TANG Ke-ke;YI Zhi-biao;XU Wei-shan;LI Zhi-xiao(Dongguan Institute of Guangzhou University of Chinese Medicine,Dongguan,Guangdong 523808;Zisun Chinese Pharmaceutical Co.Ltd.,Guangzhou,Guangdong 511430;University of Chinese Academy of Sciences,Beijing 100049,China;Shenyang Institute of Computing Technology,Chinese Academy of Sciences,Shenyang 110168)

机构地区：[1]东莞广州中医药大学研究院,广东东莞523808 [2]广州至信中药饮片有限公司,广东广州510430 [3]中国科学院大学,北京100049 [4]中国科学院沈阳计算技术研究所,沈阳110168

出　　处：《按摩与康复医学》2023年第6期48-51,共4页Chinese Manipulation and Rehabilitation Medicine

基　　金：国家重点研发计划(2017YFC1701106);东莞市社会科技发展(重点)项目(2019507101164);番禺区产业人才项目(2021-R01-6)。

摘　　要：目的:基于卷积神经网络算法建立一种能够快速、高效的识别出陈皮以及广陈皮的方法。方法:在体视镜下采集样品的外果皮图像并建立数据集。采用Imagenet预训练权重作为预训练模型,基于VGG、Inception、Resnet、DenseNet骨干模型进行训练,识别及比较,选取最优的模型进行运作。结果:Resnet与Inception网络对数据具有较好的适应性,而VGG与DenseNet存在一定的过拟合情况。通过对Resnet与Inception进行交叉验证,结果显示Resnet的准确率均高于Inception,最终选择Resnet建立一套有效识别陈皮、广陈皮的方法。结论:卷积神经网络可有效提取图像高级特征,能较为准确鉴别出陈皮与广陈皮。Objective To establish a fast and efficient method based on artificial intelligence technology to identify Chenpi and Guangchenpi.Methods:Images of Chenpi and Guangchenpi were collected by stereopicroscope.Imagenet pre-training weights were used as the pre-training model,and the classical backbone models of VGG,Inception,Resnet and DenseNet were used for training,recognition and comparison,and the optimal model was selected for establishing the recognition model.Results:Resnet and Inception have good adaptability to the data,while VGG and DenseNet have certain overfitting.Through cross-validation of Resnet and Inception,the results show that the accuracy of Resnet is higher than that of Inception.Resnet is selected as the final solution to establish a set of effective methods for identifying Chenpi and Guangchenpi.Conclusion:Convolutional neural network can effectively extract the advanced features of the image,and can accurately identify Chenpi and Guangchenpi.

关 键 词：陈皮 广陈皮 图像识别 卷积神经网络 深度学习 

分 类 号：R284[医药卫生—中药学]