Hyperspectral image-based spare autoencoder network for TVB-N measurement in pork

Abstract

Abstract. Total volatile basic nitrogen (TVB-N) content is an important index used to evaluate the freshness of pork. This research aims to develop a strategy for measurement of TVB-N content in pork through hyperspectral imaging (HSI). Spectral feature was obtained from the hyperspectral image after determining the region of interest. Nine feature wavelengths were selected using wavelength selection methods. Spare autoencoder network (SAE) was applied to obtain the internal structure of nine wavelengths. Principal component analysis (PCA) was utilized to reduce the dimension of fusion feature which is integrated with feature using SAE and the selected wavelength. A calibration model was established using least-squares support vector machine to predict TVB-N values. The correlation coefficients of prediction (RP) obtained through major components was 0.884, and its root-mean-square error of prediction was 2.93mg/100g. The residual prediction deviations (RPD) based on fusion feature was 2.14. Results demonstrated that the proposed model based on SAE-PCA exhibited potential for nondestructive detection of TVB-N content in pork.