Abstract

Food quality and safety oversight now urgently require the identification of similar food samples. Using infant rice cereals as an example, this study collected the Raman spectra of samples from various brands and used a statistical control chart method to reveal that samples from the same brand have controllable quality fluctuations. Additionally, samples from different brands also fall within the control limit, indicating a high degree of similarity between the samples. Under conditions of raw data, the extreme learning machine algorithm and Raman spectroscopy for sample identification demonstrated a recognition rate of 77.6%, suggesting that the machine learning algorithm has some identification effect. After optimizing the conditions, it was found that the recognition accuracy was significantly improved, reaching 99.9%, based on coif3 wavelet denoising and 0~0.1 normalization processing. During the experiment, it took 100 seconds to collect the Raman spectrum signal of a single sample and running the intelligent algorithm only took less than one second to obtain the calculation results. The optimization and identification methods proposed in this work have the advantages of efficiency and accuracy, which can provide a reference for the identification of similar samples.

中文翻译：

---

基于拉曼光谱结合极限学习机算法的婴儿米粉鉴别

摘要

食品质量安全监管现在迫切需要对类似食品样品进行鉴定。本研究以婴儿米粉为例，采集了不同品牌样品的拉曼光谱，并利用统计控制图方法揭示了同一品牌样品的质量波动可控。此外，不同品牌的样品也都在控制限之内，表明样品之间具有高度的相似性。在原始数据条件下，极限学习机算法和拉曼光谱对样本进行识别，识别率达到77.6%，表明机器学习算法具有一定的识别效果。优化条件后发现，基于coif3小波去噪和0~0.1归一化处理，识别准确率明显提高，达到99.9%。实验过程中，采集单个样品的拉曼光谱信号需要100秒，运行智能算法只用了不到1秒就得到了计算结果。本工作提出的优化识别方法具有高效、准确的优点，可为类似样品的识别提供参考。