Quantitative detection of chicken adulteration in lamb by VIS-NIR hyperspectral imaging with characteristic...

Meat adulteration poses a significant challenge to food authenticity and consumer protection. In this study, visible and near-infrared (VIS-NIR) hyperspectral imaging (HSI) was used for the quantitative detection of chicken adulteration in lamb. A total of 150 adulterated samples at five levels (10%, 20%, 30%, 40%, and 50%) and 30 pure lamb samples were prepared. Representative spectra were extracted using manual and automatic region-of-interest strategies. Seven characteristic wavelengths were selected using the successive projections algorithm (SPA), and their relevance for quantitative modeling was further supported by Spearman monotonic correlation analysis (SMCA). A SPA-based convolutional neural network model with channel attention and multi-scale feature enhancement (SPA-CNN-AM) was constructed by combining spectral data enhancement methods with a CNN model. The model achieved a coefficient of determination for prediction (R_P²) of 0.9967 and a root mean square error of prediction (RMSEP) of 0.0191. Furthermore, when 30% of the target-domain calibration pool was incorporated for adaptation, the SPA-CNN-AM model combined with the regularized domain transfer (RDT) strategy maintained stable predictive performance, with a concordance correlation coefficient (CCC) of 0.9716 and R_P² of 0.9428. These findings demonstrate that VIS-NIR hyperspectral imaging provides an effective spectral approach for rapid and non-destructive quantitative analysis of lamb adulteration and shows potential for cross-species external predictive transfer under limited target-domain adaptation.