Rapid and chemical-free technique based on hyperspectral imaging combined with artificial intelligence for monitoring quality and shelf life of dried shrimp.

Abstract

A rapid and chemical-free method based on hyperspectral imaging (HSI) integrated with artificial intelligence (AI) for monitoring dried shrimp quality was developed. Dried shrimp was packaged in a polypropylene bag and chronologically monitored for changes in quality indices, including astaxanthin, thiobarbituric acid reactive substance (TBARS), total volatile basic nitrogen (TVB-N), and sensory attributes during storage at 30 °C. A hyperspectral imaging system was operated in the visible to near infrared regions (400-1000 nm) to acquire the hypercube data of the dried shrimp. The obtained spectral data can be recognized as fingerprints ascribing to the changes in each quality index. The reflectance spectra were mathematically pretreated using a moving average smoothing technique to eliminate the variability caused by non-uniformity of light scattering prior to further data analyses. Principal component analysis (PCA) was used as an unsupervised machine learning to explore the relationships among wavelength variables and to reduce the dimensionality of the hyperspectral data. Feed forward with back propagation artificial neural network (ANN) was then used to simultaneously evaluate astaxanthin, TBARS, TVB-N and sensory characteristic of dried shrimp and gave coefficient of determination of 0.96, 0.95, 0.92, and 0.93 and root mean squared error of 1.12 μg/g, 0.22 mg MDA/kg, 3.19 mg N/100 g and 0.68 score, respectively. Pseudo-color distribution maps were also constructed from the ANN-predicted values to facilitate the data visualization of time-associated quality changes in dried shrimp.  The results suggested that HSI integrated with artificial intelligence is effective to evaluate quality and shelf life of dried shrimp during storage.