Optimized identification of cheese products based on Raman spectroscopy and an extreme learning machine

Abstract

Rapid and intelligent identification of similar samples is an important technical aid for food quality and safety assessment. At present, relevant research studies are still relatively scarce. In this paper, Raman spectra of different brands of cheese products with different integration times were collected and analyzed statistically. The research showed that the original Raman spectral data of different brands of cheese products had a high similarity. The conventional statistical control chart method revealed that the samples of the same brand showed certain quality fluctuations, but it was still difficult to achieve effective differentiation from other brand samples. Different spectral preprocessing methods had a greater impact on the classification accuracy of the intelligent extreme learning machine algorithm, and different spectral integration times also had a greater impact. Before optimization, the intelligent recognition algorithm had only about 40% average accuracy with the integration time of 10 s. Under the optimal conditions – an 80 s Raman spectral integration time, wavelet denoising (coif wavelet basis, 5 decomposition layers), normalization processing ([−1, 1] interval), principal component dimensionality reduction (74 principal components were extracted, which can represent 100% of the information of the original data) and 400 hidden-layer neurons – the average recognition accuracy was 98%, and the algorithm operation time was less than 5 s. The establishment of this method provides a technical reference for rapid discrimination in food quality control.