Nondestructive quantifying total volatile basic nitrogen (TVB-N) content in chicken using hyperspectral imaging (HSI) technique combined with different data dimension reduction algorithms

Abstract

Hyperspectral imaging (HSI) system has been used to assess the chicken quality in this work. Principle component analysis (PCA) and Ant Colony Optimization (ACO) were comparatively used for data dimension reduction. First, we selected 5 dominant wavelength images from chicken hypercube using PCA and ACO. Then, 6 textural variables based on statistical moments were extracted from each dominant wavelength image, thus totaling to 30 variables. Next, we selected the classic back propagation artificial neural network (BPANN) algorithm for modeling. Experimental results showed the performance of ACO-BPANN model is superior to that of PCA-BPANN model, and the optimum ACO-BPANN model was achieved with RMSEP=6.3834 mg/100g and R=0.7542 in the prediction set. Our work implies that HSI integrating spectral and spatial information has a high potential in quantifying TVB-N content of chicken in rapid and non-destructive manner, and ACO has superiority in dimension reduction of hypercube.

Keywords: ACO algorithm; Chicken spoilage; Hyperspectral imaging (HSI); Texture analysis; Wavelength selection.