基于卷积神经网络的烟丝物质组成识别方法

摘要

为了满足混合烟丝物质组成的检测需求,针对叶丝、梗丝、膨胀叶丝、再造烟叶丝4种烟丝组成的结构特征差异,采用深度学习的方法,建立基于卷积神经网络的烟丝组成识别模型,使用体现烟丝微观结构特征的局部特征图片作为神经网络的输入,分析识别出每个特征图片对应的输出结果,通过统计方法得出烟丝的组成成分.结果表明:识别模型对烟丝训练样本和测试样本的识别正确率分别为100%和84.95%,模型中的卷积神经网络结合相应的结果表示方法,可以更好地提高烟丝样本成分的识别正确率.%For evaluating the blending consistence in cigarette production, the make-up of cut filler, including cut strips, cut stems, expanded cut strips and cut reconstituted tobacco, was identified with a model based on convolutional neural network incorporating deep learning approach. The local images, which reflected the microstructural characteristics of cut filler were used as the inputs of neural network. The output corresponding to each local characteristic image was analyzed and identified, and via statistical analysis the constituents of cut filler were determined. The results showed that the identification accuracies of the model for training samples and testing samples were 100% and 84.95%, respectively. The convolutional neural network combining with the corresponding method of result expressing in the model effectively promotes the identification accuracy for cut filler samples.