Deep learning-assisted fluorescence spectroscopy for food quality and safety analysis

文献类型： 外文期刊

第一作者： Yuan, Yuan

作者： Yuan, Yuan;Xu, Qian;Ji, Zengtao;Fan, Yanwei;Shi, Ce;Ji, Zengtao;Fan, Yanwei;Shi, Ce;Ji, Zengtao;Fan, Yanwei;Shi, Ce;Yuan, Yuan;Xu, Qian;Lyu, Jian;Ertbjerg, Per

作者机构：

关键词： Fluorescence spectroscopy; Limitations of machine learning; Classification and regression; Model optimization; Model performance evaluation; Application

期刊名称：TRENDS IN FOOD SCIENCE & TECHNOLOGY （ 影响因子：15.4； 五年影响因子：18.4 ）

ISSN： 0924-2244

年卷期： 2025 年 156 卷

页码：

收录情况： SCI

摘要： Background: Fluorescence spectroscopy has been widely employed in the quality assessment of food and agricultural products due to its rapid and accurate measurement characteristics. The large amount of fluorescence data or images generated by fluorescence spectroscopy requires more efficient chemometric methods to process and analyze them. However, conventional machine learning models struggle to achieve high-precision predictions when analyzing high-dimensional fluorescence data samples. Deep learning algorithms exhibit powerful automatic learning capabilities in feature extraction and regression modeling of fluorescence spectra. Scope and approach: The complex, abstract and high-dimensional features of fluorescence spectroscopy are firstly demonstrated through the characterization of fluorescent substances in food products. Secondly, this paper highlights various challenges confronting the fluorescence spectrum analysis process and summarizes several deep learning algorithms that can address these solutions, including the convolutional neural network (CNN), long and short-term memory network (LSTM), and auto encoder (AE). Additionally, the application of deep learning models based on fluorescent data in food detection is reviewed in this article according to different testing objectives, including food safety inspections, food quality assessment, adulteration identification, and variety identification. The review also focuses on the future development trend of this technique in food quality and safety detection.
Key findings and conclusions: Deep learning approaches combined with fluorescence spectroscopy exhibits immense potential in food quality detection and food discrimination classification. The selections of representative input parameters, suitable preprocessing methods and optimization methods can effectively tackle the problems of lack of samples and model over-fitting. Owing to the rapid advancement of artificial intelligence, the deep learning-based fluorescence spectroscopy technology is poised to evolve towards high precision, high throughput, automation and cost-effectiveness.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序