Rapid and non-destructive detection of shrimp freshness based on multi-level data fusion strategies of hyperspectral and Raman spectroscopy using a spoilage gas-responsive chip

Shrimp quality deteriorates during storage, making the assessment of freshness vital in the food industry. In this study, a spoilage gas-responsive chip was fabricated to integrate hyperspectral imaging and Raman spectroscopy, enabling the construction of a multimodal nondestructive detection model based on a multi-level data fusion strategy. Three different levels of fusion strategies were analyzed and compared, namely raw-data spectrum fusion, refined-feature spectrum fusion, and cross-model decision fusion. The results demonstrated that the cross-model decision fusion strategy employing adaptive Bayesian weighting for decision-making achieved excellent performance, with Rp 2 , RMSEP, and RPD values of 0.9911, 0.9056, and 10.3730, respectively. In subsequent external validation, the Rv 2 and RMSEV were 0.9352 and 2.1479, respectively, demonstrating significant improvement over the single-modal model and the spectrum fusion models. This study offers a promising approach for rapid, nondestructive freshness detection during shrimp storage.