基于改进深度学习模型IRCNN的卷烟真伪鉴别
    点此下载全文
引用本文:李海燕,李郸,马慧宇,肖燕.基于改进深度学习模型IRCNN的卷烟真伪鉴别[J].计算技术与自动化,2023,(1):188-192
摘要点击次数: 366
全文下载次数: 0
作者单位
李海燕,李郸,马慧宇,肖燕 (云南省烟草质量监督检测站云南 昆明 650104) 
中文摘要:针对常用的卷烟包装图像真伪鉴别,目前深度学习方法需要较高的设备成本与较长的训练时间。本文提出了一种基于Inception和ResNet卷积神经网络结合的卷烟包装图像真伪鉴别模型IRCNN(Inception-ResNet Convolutional Neural Network)。利用Inception网络并行结构自动学习并提取卷烟包装图像的不同尺度特征,同时在线路中加入三维卷积核,有效地增强不同线路之间的信息交互。利用残差结构减少由于网络加深导致的模型退化。实验结果表明,与其他深度学习方法相比较,本文提出的方法不仅减少算法设备成本和训练时间,而且准确率可达到99.88%。因此,通过采用多线路Inception和残差网络相结合的IRCNN模型,可以有效地提高卷烟真伪鉴别效率和精度,为将来实际应用提供技术支持。
中文关键词:卷烟包装  真伪鉴别  卷积神经网络  Inception  ResNet
 
Cigarette Authenticity Identification Based on Improved Deep Learning Model IRCNN
Abstract:At present, the deep learning method commonly used for cigarette packaging image authenticity identification needs high equipment cost and long training time. This paper proposes a cigarette packaging image authenticity identification model IRCNN (Inception-ResNet Convolutional Neural Network) based on the combination of Inception and ResNet Convolutional Neural Network. By using the Inception network’s parallel structure, the network automatically learns and extracts features of different scales of cigarette images, resizing structures are introduced to reduce degradation caused by network deepening, and three-dimensional convolution is added to effectively enhance information between different channels to blend. The experimental results show that, compared with other deep learning methods, the proposed method not only reduces the equipment cost and training time, but also achieves an accuracy of 99.88%. Through the combination of multi-circuit Inception and residual network, the efficiency and accuracy of cigarette authenticity identification can be effectively improved, providing technical support for future practical application.
keywords:cigarette packaging  authenticity identification  convolutional neural network  Inception  ResNet
查看全文   查看/发表评论   下载pdf阅读器