Nondestructive Freshness Assessment of Litopenaeus vannamei Based on Polarization Imaging and Deep Learning

ABSTRACT Freshness is a key determinant of the quality and commercial value of Litopenaeus vannamei ( L. vannamei ). This study proposes a freshness assessment approach that integrates polarization imaging with deep learning for the identification of L. vannamei freshness. Four‐angle polarization images were acquired from samples stored at 4°C for 0–5 days, and the depolarization degree (DoD) and angle of polarization (AoP) were computed. With increasing storage duration, the overall DoD of the shrimp increased, while the AoP distribution became more dispersed, indicating progressive disruption of muscle fiber microstructure and protein degradation. A ResNet‐18‐based multi‐input model was developed to compare feature representations across different modalities. Among the evaluated inputs, the four‐angle polarization images demonstrated the best classification performance on the Nantong dataset, with accuracy, precision, recall, and F1‐score all exceeding 97%. These results provide a new solution for the rapid and non‐destructive quality inspection of aquatic products.