Nondestructive Detection of Glucocorticoid Residues in Chicken Meat Using Hyperspectral Imaging

ABSTRACT Residual glucocorticoids in chickens can harm human health, and traditional detection methods can cause environmental pollution. In this study, hyperspectral imaging (HSI) technology was used to investigate the feasibility of nondestructively identifying chicken samples spiked with five different glucocorticoids. First, regions of interest (ROIs) were used to extract hyperspectral data. After preprocessing, the obtained average hyperspectral data were transformed at eight scales using multiscale continuous wavelet transform (CWT), and distributed stochastic neighbor embedding (t‐SNE) and decision tree (DT) models were used to select the optimal transformation scale. Among the four feature extraction methods, the combination of stacked autoencoders and Fisher score demonstrated the best performance for characterizing the spectral features of the treated chicken samples. Finally, K ‐nearest neighbors (KNNs), support vector machine (SVM), Light Gradient Boosting Machine (LightGBM), and DT were selected as classifiers to evaluate the overall performance. The results showed that the CWT‐SAE‐FisherScore‐SVM model achieved the best classification performance (accuracy = 99.76%). Compared to benchmark methods such as PCA + SVM, this method demonstrates greater robustness and clearer physical significance at extremely low feature dimensions. This study established a proof‐of‐concept rapid screening framework for identifying simulated glucocorticoid residues in chicken in experimental settings. Although this model currently focuses on artificially spiked samples, it provides a robust technical foundation and an automatic segmentation algorithm to support the future development of real‐time intelligent monitoring systems for practical food safety applications.