Quantitative analysis and visualization of chemical compositions during shrimp flesh deterioration using hyperspectral imaging: A comparative study of machine learning and deep learning models.
Journal:
Food chemistry
PMID:
40174377
Abstract
The current work explores hyperspectral imaging (HSI) to quantitatively identify changes in TVB-N and K value during shrimp flesh deterioration. The work developed low-level data fusion (LLF) and predictive models using both machine learning methods (PLS) and deep learning methods (CNN, LSTM, CNN-LSTM). Results indicate that deep learning methods show comparable performance due to their superior feature extraction and fitting capabilities, but traditional chemometric methods outperform deep learning models, achieving R = 0.9431 (TVB-N), and R = 0.9815 (K value). Subsequently, spatial distribution maps were generated based on the optimal predictive models to visualize the chemical composition changes in shrimp flesh. This approach allows for rapid, non-destructive prediction of spoilage-related changes. This technology can monitor shrimp quality in cold chain logistics, improve inventory management, and ensure seafood quality. Future research should optimize models for varied conditions and explore combining HSI method with other sensor technologies to enhance shrimp quality evaluation comprehensively and accurately.