AI-driven non-destructive identification of expired medicines using hyperspectral imaging and FTIR spectroscopy for pharmaceutical quality assessment

Ensuring pharmaceutical quality and safety is essential to prevent the consumption of expired medicines and associated health risks. This study explores an AI-driven, non-destructive framework that integrates hyperspectral imaging (HSI) and Fourier Transform Infrared (FTIR) spectroscopy to classify fresh and expired medicines. FTIR spectroscopy served as a supportive qualitative chemical characterization, analyzing variations in functional groups and molecular vibrations between fresh and expired medicines. A complementary HSI dataset was acquired with the Specim FX10 camera in the 400–1000 nm VNIR range. Spectral preprocessing included ROI-based feature extraction, Empirical Line Method (ELM) calibration for radiometric correction, and Savitzky–Golay filtering to remove noise and smooth the medicine spectral signature. Machine learning models, including Cat Gradient Boosting (CatBoost), Histogram-Based Gradient Boosting (HGB), Light Gradient Boosting Machine (LightGBM), and Extreme Gradient Boosting (XGBoost), were used to evaluate classification performance. Experimental evaluation across six pharmaceutical classes demonstrated strong classification performance. Among the evaluated models, the LightGBM model achieved 100% training accuracy and 99.92% validation accuracy, with precision, recall, and F1-score of 99.95%. It also achieved 99.94% for both the Matthews Correlation Coefficient (MCC) and Cohen’s kappa, a Jaccard Index of 99.90%, and a 10-fold cross-validation accuracy of 99.92% ± 0.053% (mean ± standard deviation). In addition, the model produced very low Log Loss (0.00524) and Hamming Loss (0.00050), indicating very good classification performance in distinguishing fresh and expired medicines, together with high predictive confidence and a very low misclassification rate. The proposed AI-driven HSI-based framework demonstrates reliable and efficient performance for rapid, non-destructive identification of expired medicines and pharmaceutical quality assessment. The findings demonstrate the strong potential of integrating spectroscopic techniques with artificial intelligence for pharmaceutical quality inspection.