Feature wavelength selection and model development for rapid determination of myoglobin content in nitrite-cured mutton using hyperspectral imaging

Abstract

Hyperspectral imaging (HSI), a non-destructive method combining image and spectral techniques to obtain image and spectral information of target objects, and qualitative and quantitative analyses using spectral data, has been widely used in agricultural product detection. This study was conducted to assess the feasibility of using HSI (900–1700 nm) for rapid determination of myoglobin contents (DeoMb, OxyMb and MetMb) in nitrite-cured mutton during refrigerated storage. The data were analyzed using evolutionary computing methods, including partial least squares regression (PLSR) and least squares support vector machines (LSSVM). Corresponding feature wavelengths were then selected by using competitive adaptive reweighted sampling (CARS), variable combination population analysis (VCPA) and interval variable iterative space shrinkage approach (iVISSA) respectively. The simplified iVISSA-PLSR model showed the best performance for determining DeoMb with Rp of 0.9017 and RMSEP of 2.3509. In addition, the best model for OxyMb was CARS-PLSR, with Rp of 0.9661 and RMSEP of 2.3762. Moreover, the CARS-LSSVM model showed a better performance with a higher Rp (0.8931) and a lower RMSEP (3.2743) for MetMb. Furthermore, the overall performance of wavelength selection methods in terms of CARS (RpDeoMb ≥ 0.8943, RpOxyMb ≥ 0.9607, RpMetMb ≥ 0.8910) and iVISSA (RpDeoMb ≥ 0.8976, RpOxyMb ≥ 0.9568, RpMetMb ≥ 0.8880) were equivalent and slightly better than VCPA (RpDeoMb ≥ 0.8711, RpOxyMb ≥ 0.9573, RpMetMb ≥ 0.8808). The results demonstrated that effective wavelength selection method can improve the performance of HSI system for the determination of myoglobin contents in comparison to full wavelength.