Classification of broiler breast fillets based on multi-spectral fusion of NMR, FTIR and fluorescence

To address the problem of insufficient information dimension of single spectroscopic techniques in broiler wooden breast (WB) grading, and to provide a methodological basis for the development of efficient and objective industrial grading technology, this study established a three-level grading method for broiler breast fillets based on multi-spectral fusion of low-field nuclear magnetic resonance (LF-NMR), Fourier transform infrared (FTIR) spectroscopy, and three-dimensional excitation-emission matrix (EEM) fluorescence spectroscopy. A total of 150 breast fillets from 42-day-old male Arbor Acres broilers, categorized into normal breast (NORM), moderate WB (MOD), and severe WB (SEV), were investigated. Fourteen data combinations (7 pure spectral and 7 full-information fusion incorporating basic physicochemical indicators) were constructed via the low-level data fusion (LLDF) strategy, and three-class classification models were developed using partial least squares discriminant analysis (PLS-DA), support vector machine (SVM), and multilayer perceptron (MLP) algorithms. The results showed that the water distribution, protein structure, and oxidative status of WB exhibited significant gradient changes with increasing severity; FTIR was the optimal single-modal spectroscopic technique, and the SVM model achieved the best overall performance. Notably, the combination of tri-spectral fusion and basic physicochemical indicators achieved 100% test set classification accuracy across all models. This study provides a reliable technical solution for the rapid grading of WB in the poultry industry, balancing detection accuracy, efficiency and practical industrial feasibility.