|
|
|
| Construction of a Predictive Model for the Determination of Salami Sausage Maturity Based on Near Infrared Spectroscopy and Machine Learning |
| CAI Min, LIU Yuhao, JIAO Yushan, TANG Wensheng, LIU Yingli, YANG Yi, WANG Jun, YANG Li |
| 1. Key Laboratory of Geriatric Nutrition and Health, Ministry of Education, Beijing Technology and Business University, Beijing 100048, China; 2. School of Computer and Artificial Intelligence, Beijing Technology and Business University, Beijing 100048, China; 3. Key Laboratory of Meat and Meat Products Regulation Technology, State Administration for Market Regulation, Shandong Institute for Food and Drug Control, Jinan 250101, China; 4. Yantai Research Institute, China Agricultural University, Yantai 264670, China |
|
|
|
|
Abstract In this study, the effects of different starter cultures on the physicochemical quality of salami during ripening were systematically evaluated using near infrared spectroscopy (NIRS) and machine learning (ML) algorithms, and a nondestructive method for determining the key ripening-related physicochemical indexes of salami was established using NIRS and ML. Experimental groups with different mixed starter cultures were set up (combinations of Pediococcus pentosus, Lactobacillus casei, Leuconostoc intestinalis, Staphylococcus xylosus). The physicochemical properties of fermented sausage including pH, moisture content (MC), water activity (aw), color difference (ΔE), and resilience index (RI) were dynamically monitored. Three predictive models were developed using partial least squares regression (PLSR), random forest (RF) and extreme learning machine (ELM) based on the NIRS data. The predictive ability of three data input methods, namely, the full-band data, principal component analysis (PCA) and regression coefficient (RC), was comparatively analyzed. The results showed that 1) the sausage made with the P. pentosus M10 + L. casei M13 + L. intestinalis + S. xylosus received the highest sensory score with outstanding performance in color, texture, flavor and overall acceptability; 2) NIRS allowed the prediction of aw, MC, ΔE and RI, while the prediction accuracy for pH was low (coefficient of determination of prediction set (Rp 2 ) = 0.45); 3) both feature extraction methods effectively reduced data dimension; the cumulative contribution rate of the first five principal components of PCA was 98.86%, while RC reduced the number of spectral variables from 176 to 10–25; and 4) the PLSR-RC model was the optimal predictive model for aw with Rp 2 of 0.92 and mean squared error of prediction (MSEP) of 0.02, the ELM-full band model was the optimal predictive model for MC (Rp 2 = 0.92, MSEP = 2.56), and the PLSR-full band model was the optimal predictive model for both ΔE and RI with Rp 2 of 0.71 and 0.83, respectively. In summary, the mixedstrain starter can significantly improve the physicochemical quality and sensory characteristics of sausage, and the PLSR model shows the best prediction accuracy and stability in the analysis of NIRS data, providing reliable technical support for the rapid detection of fermented meat maturity.
|
|
|
|
|
|
|
|
| [1] |
FENG Yujian, WANG Hongwei, XU Na, PANG Jianlong, GUO Zhifeng, ZHANG Qingyong, YAN Ruiping, ZHENG Zhaoqin, LI Feifeng, ZHANG Li, Lü Qingqin, WANG Peng. Effect of Boiling and Steaming on the Flavor Quality of Grilled Chicken Rack: A Comparative Study by Gas Chromatography-Ion Mobility Spectrometry and Chemometric Analysis[J]. Meat Research, 2026, 40(9): 70-79. |
|
|
|
|
2026, Vol. 40 
