Prediction and Spatial Visualization of Water Holding Capacity in Frozen Lettuce via Mueller Matrix Imaging and Machine Learning

Abstract

The water holding capacity (WHC) of complex food systems, such as lettuce during freezing, is governed by microstructural integrity. Optically quantifying this integrity, however, remains a challenge, as conventional optical methods typically probe chemical signatures rather than the underlying physical architecture. This study introduces a snapshot Mueller matrix polarimetric imaging system, employing quad-channel parallel demodulation, to directly assess microstructural changes and track the dynamic evolution of WHC in frozen lettuce. By acquiring full Mueller matrices and leveraging machine learning, we established quantitative models to predict WHC from 18 derived polarimetric features. The random forest regression (RFR) model yielded the highest predictive accuracy on an independent test set (R² = 0.8718, RMSE = 0.0356). Feature importance analysis confirmed that parameters reflecting tissue anisotropy (m11, m22) and disorder (depolarization, Δ) were the most critical predictors, establishing a direct link between the optical measurement and physical degradation. Pixel-wise mapping visualized the spatio-temporal evolution of WHC, revealing a transition from initial, heterogeneous damage to widespread structural collapse. Notably, the system’s parallel architecture acquires the four analysis-state images in a single snapshot for each illumination state. This synchronous demodulation provides inherent data consistency and represents a simplified design compared to sequential analysis schemes. This research establishes Mueller matrix polarimetry as a powerful paradigm for optical food quality inspection. By directly correlating optical signatures with microstructural integrity, it also demonstrates significant potential for intelligent online monitoring in agricultural product processing.