Discrimination of shelled shrimp (Metapenaeus ensis) among fresh, frozen-thawed and cold-stored by hyperspectral imaging technique

The potential of a Vis-NIR (400-1000 nm) hyperspectral imaging system was investigated to discriminate fresh (F), cold-stored (C-S) and frozen-thawed (F-T) shelled shrimp (Metapenaeus ensis), to detect or prohibit illegally substituted and mislabeled products from the market. In this study, eight feature wavelengths (783, 689, 435, 416, 813, 639, 452 and 478 nm) were extracted by uninformative variable elimination (UVE) based on partial least squares (PLS) with successive projections algorithm (SPA) from the whole variables. Meanwhile, grey level co-occurrence matrix (GLCM) was applied to extract the first three principal components (PCs) images, which explained more than 99% of variances of all spectral bands. The spectral and textural data were combined for discrimination. Random forest (RF) and soft independent modeling of class analogy (SIMCA) classifiers were utilized to discriminate F, C-S and F-T shrimp based on: (1) whole spectral data, (2) optimal spectral data, (3) textural data, and (4) fusion of textural and optimal spectral data. Based on (4), satisfying results were obtained using RF and SIMCA with correct classification rates (CCRs) of 91.11% and 88.89% in the prediction sets, respectively, confirming the feasibility of hyperspectral imaging to classify fresh shrimps and those after cold storing or freezing.