基于激光诱导击穿光谱结合深度学习的鲜肉快速识别方法研究

doi:10.3964/j.issn.1000-0593(2025)12-3317-07

摘要
参考文献
相关文章 (15)

全文: PDF (10705 KB)
输出: BibTeX | EndNote (RIS)

摘要：肉类是人类饮食中重要的蛋白质和营养物质来源，而肉类食品掺假问题已成为我国食品安全领域的重大问题，针对传统肉类食品检测方法存在操作复杂、程序耗时、设备造价高昂的问题，提出利用激光诱导击穿光谱技术（LIBS）结合深度学习网络模型的方法对多品种肉类组织进行快速检测与分类识别。通过LIBS技术分别采集了牛肉、羊肉、猪肉3种肉类光谱数据，选取5种元素9个波段谱线作为分析谱线及模型输入量进行建模识别，设计了1种以ResNet18为主干网络，同时设计3种机器学习模型，对光谱数据进行建模识别。结果表明，深度学习网络模型识别能力最优，精确度达到了98.1%。120组光谱数据中，牛肉、羊肉、猪肉光谱数据分别有117、119、117组光谱数据识别正确，在与同为深度学习网络模型的横向对比中，ResNet18网络模型对肉类光谱数据的识别能力优于GoogLeNet、Vgg16、ResNet50的3种深度学习模型，在此基础上通过重新采集的数据对模型泛化性进行验证，精度达到98.9%，说明模型能够在较高识别精度的基础上，保持较强的跨数据集识别能力，具有良好的泛化性与一致性。以上研究表明，通过LIBS与卷积神经网络结合的方式能够在多品种肉类分类识别任务中提供不同品种肉类客观量化差异信息，并具有快速和原位诊断不同种类肉类组织的潜力。

关键词：激光诱导击穿光谱；多品种肉类识别；深度学习

Abstract：Meat is an important source of protein and nutrients in the human diet, and the adulteration of meat has become a major problem in the field of food safety in China. To address the problems of complex operations, time-consuming procedures, and high equipment costs in traditional meat detection methods, this paper proposes using Laser-induced Breakdown Spectroscopy (LIBS) combined with a deep learning network to detect and classify multi-variety meat tissues quickly. Spectral data for beef, mutton, and pork were collected using LIBS. Nine spectral lines of five elements were selected as the analysis spectral lines and model input for modeling and recognition. A ResNet18 backbone network was designed, and three machine learning models were designed to model and recognize the spectral data.The results show that the deep learning network achieves the best recognition performance, with an accuracy of 98.1%. Among the 120 groups of spectral data, 117, 119, and 117 groups of beef, mutton, and pork spectral data were identified correctly, respectively. In the horizontal comparison using the same deep learning model, the ResNet18 model was superior to the three deep learning models, GoogLeNet, Vgg16, and ResNet50, in the recognition of meat spectral data. On this basis, the model's generalization was verified using re-collected data, and the accuracy reached 98.9%, indicating that the model maintains strong cross-data-set recognition ability. It has good generalization and consistency. The above research shows that the combination of LIBS and a convolutional neural network can provide objective, quantitative information on differences between meat varieties in multi-variety meat classification and recognition tasks and has the potential to quickly and in situ diagnose different types of meat tissue.

Key words：Laser-induced breakdown spectroscopy; Multi-species meat identification; Deep learning

收稿日期: 2025-03-31 修订日期: 2025-09-19

中图分类号:

O433.4

基金资助: 国家自然科学基金项目（61575030），国家重大科学仪器设备开发专项（2014YQ12035104），吉林省科学技术厅项目（20240304014SF，20220203178SF，20200403008SF，20180201046YY), 吉林省教育厅项目JJKH20241147KJ)，省发改委项目（2018C034-3）资助

通讯作者: 方健 E-mail: 1419404982@qq.com

作者简介: 孙浩然，1992年生，吉林交通职业技术学院讲师 e-mail: 1515754043@qq.com

引用本文:

孙浩然，王思文，赵春园，林晓梅，高勋，方健. 基于激光诱导击穿光谱结合深度学习的鲜肉快速识别方法研究[J]. 光谱学与光谱分析, 2025, 45(12): 3317-3323.
SUN Hao-ran, WANG Si-wen, ZHAO Chun-yuan, LIN Xiao-mei, GAO Xun, FANG Jian. Rapid Identification of Fresh Meat Based on Laser-Induced Breakdown Spectroscopy Combined With Deep Learning Methods. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(12): 3317-3323.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2025)12-3317-07 或 https://www.gpxygpfx.com/CN/Y2025/V45/I12/3317