Non-destructively sensing pork’s freshness indicator using near infrared multispectral imaging technique

•Multispectral imaging system based on near infrared bands was developed.•Pork freshness was successfully evaluated by NIR multispectral imaging.•A novel AdaBoost + BPANN algorithm was used for model calibration.

Total volatile basic nitrogen (TVB-N) content is one of core indicators for evaluating pork’s freshness. This paper attempted to non-destructively sensing TVB-N content in pork meat using near infrared (NIR) multispectral imaging technique (MSI) with multivariate calibration. First, a MSI system with 3 characteristic wavebands (i.e. 1280 nm, 1440 nm and 1660 nm) was developed for data acquisition. Then, gray level co-occurrence matrix (GLCM) was used for characteristic extraction from multispectral image data. Next, we proposed a novel algorithm for modeling-back propagation artificial neural network (BP-ANN) and adaptive boosting (AdaBoost) algorithm, namely BP-AdaBoost, and we compared it with two commonly used algorithms. Experimental results showed that the BP-AdaBoost algorithm is superior to others with the root mean square error of prediction (RMSEP) = 6.9439 mg/100 g and the correlation coefficient (R) = 0.8325 in the prediction set. This work sufficiently demonstrated that the MSI technique has a high potential in non-destructively sensing pork freshness, and the nonlinear BP-AdaBoost algorithm has a strong performance in solution to a complex data processing.