基于太赫兹及含Inception块残差网络的北柴胡鉴别  

Identification of Bupleuri chinense Based on Terahertz and Residual Network with Inception Blocks

作　　者：叶华清 蔡圣杰 郑成勇[1] YE Hua-qing;CAI Sheng-jie;ZHENG Cheng-yong(School of Mathematics and Computational Science,Wuyi University,Jiangmen 529020,China)

机构地区：[1]五邑大学数学与计算科学学院,广东江门529020

出　　处：《五邑大学学报（自然科学版）》2023年第4期47-52,共6页Journal of Wuyi University(Natural Science Edition)

基　　金：广东省教育科学规划课题(2021GXJK308);五邑大学港澳联合研发基金(2022WGALH16)。

摘　　要：为更准确鉴别藏柴胡、锥叶柴胡和北柴胡,本文首先对柴胡样本进行太赫兹光谱测定,然后用含Inception块的残差网络对光谱数据进行识别.本文方法将Inception块中的卷积核收缩成一维,通过一维Inception块堆叠及残差连接来构建残差网络的主干部分.主干部分后面依次是全局平均池化(Global Average Pooling,GAP)层、全连接(Full Connection,FC)层和Softmax层,其中,网络的主干部分用于对输入的太赫兹光谱数据进行多尺度特征提取,GAP层用于汇聚多尺度特征,FC层和Softmax层用于实现最后的分类.本文算法与9种传统模式识别算法进行了对比实验.结果表明,本文算法鉴别精度达88.99%,优于9种传统模式识别算法.本文算法为北柴胡的鉴别提供了新的解决方案.To more accurately identify Bupleurum chinense from Tibetan Bupleurum and Conical Leaf Bupleurum,the Terahertz spectra of three Bupleurum samples are collected firstly,and a residual network containing Inception blocks is then adopted to identify the spectral data.We shrink the convolutional kernels in Inception blocks into one dimension,and construct the backbone of the residual network through one-dimensional Inception blocks stacking and residual connections.The backbone is followed by the Global Average Pooling(GAP)layer,Full Connection(FC)layer,and Softmax layer.Among them,the backbone of the network is used for multi-scale feature extraction of inputted Terahertz spectral data,the GAP layer is used to aggregate multi-scale features,and the FC layer and Softmax layer are used to achieve the final classification.The proposed algorithm was compared with 9 traditional pattern recognition algorithms through experiments,the experimental results of which show that the identification accuracy of the proposed algorithm reaches 88.77%,which is superior to 9 traditional pattern recognition algorithms.This algorithm provides a new solution for the identification of Bupleurum chinense.

关 键 词：太赫兹 深度学习 北柴胡鉴别 Inception块 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]