Nondestructive evaluation of strawberry freshness and quantitative analysis of blue light irradiation effects using hyperspectral imaging and machine learning.

The rapid degradation of strawberry freshness poses a significant challenge for quality control in the supply chain. However, traditional evaluation methods are inherently destructive, time-consuming, and fundamentally subjective due to the absence of a universal gold standard, thereby hindering real-time monitoring. Here, we demonstrate a rapid and non-destructive strategy combining hyperspectral imaging (HSI) with machine learning (ML) algorithms to predict the freshness of strawberries over varying storage days. A systematic evaluation identified the MSC + SG2-GA-BPNN architecture-integrating multiplicative scatter correction (MSC) and Savitzky-Golay second derivative (SG2) for spectral preprocessing, a genetic algorithm (GA) for feature selection, and a back-propagation neural network (BPNN) for classification-as the optimal pipeline, achieving classification accuracies of 100% and 99% under 4 °C and 20 °C storage, respectively. Utilizing this validated model as a dynamic monitoring tool, we uncovered a biphasic, dose-dependent relationship between blue light exposure duration (450 nm, 40 mW cm-2) and strawberry preservation at 4 °C. Specifically, daily irradiation for 5 h significantly improved preservation efficacy, while prolonged exposure (20 h) significantly accelerated quality deterioration. These results provide a robust technical strategy for intelligent postharvest management and introduce a novel methodology for quantifying environmental effects on produce quality through high-performance predictive frameworks.