激光与光电子学进展, 2018, 55 (4): 043001, 网络出版: 2018-09-11
深度学习与激光诱导荧光在假酒识别中的应用 下载: 1254次
Application of Counterfeit Liquor Recognition Based on Deep Learning and Laser Induced Fluorescence
光谱学 荧光光谱 激光诱导荧光 假酒识别 深度学习 spectroscopy fluorescence spectra laser induced fluorescence counterfeit liquor recognition deep learning
摘要
假酒的快速识别在食品安全领域具有重要意义,但现有的白酒检测技术无法既快速又准确地识别市售假酒。提出一种快速辨识白酒真假的方法,即利用激光诱导荧光技术获取待测白酒的荧光光谱,调整其大小后输入深度学习算法,进而辨识其真假。实验酒样选取4个样本,包括3个品牌的2种度数的白酒,每种酒样采集100个荧光光谱,然后从每种酒样的100个光谱中随机选取80个用于深度学习算法模型的训练,剩余20个用于测试训练好的模型。结果表明:品牌不同、度数不同的白酒,其荧光光谱都存在明显差异。在模型测试中,4种酒样荧光光谱的识别率为98.44%。激光诱导荧光技术结合深度学习能准确识别出白酒的品牌与度数。
Abstract
The fast recognition of counterfeit liquor is significant in the field of food safety, while the existing liquor detection technologies cannot quickly identify various kinds of counterfeit liquors in the market. We propose a quick method of liquor-authenticity identifying. Firstly, we use laser induced fluorescence technique to collect fluorescence spectra of the liquor under test. Then we adjust the size of fluorescence spectra to input into the deep learning algorithm, Finally, we can identify the authenticity based on the algorithm. We select four samples, which are three liquor brands with two liquor degrees, and collect 100 fluorescence spectra for each liquor sample. Then we select 80 spectra from the 100 spectra randomly for model training deep learning algorithm. Finally, we detect the rest 20 spectra for the trained model. The experimental results show that there are significant differences of fluorescence spectra in different liquor brands and different liquor degrees. In model test, the recognition rate of fluorescence spectra in four samples is 98.44%. The results indicate that the laser induced fluorescence technology and deep learning can identify the brand and degree of liquor precisely.
来文豪, 周孟然, 王亚, 胡锋, 李大同, 赵舜. 深度学习与激光诱导荧光在假酒识别中的应用[J]. 激光与光电子学进展, 2018, 55(4): 043001. Wenhao Lai, Mengran Zhou, Ya Wang, Feng Hu, Datong Li, Shun Zhao. Application of Counterfeit Liquor Recognition Based on Deep Learning and Laser Induced Fluorescence[J]. Laser & Optoelectronics Progress, 2018, 55(4): 043001.