Accurate and non-destructive identification of origins for lily using near-infrared hyperspectral imaging combined with machine learning.

Origin identification is important for safeguarding food quality, as the content of various ingredients in a product's raw material depends on its origin. By combining near-infrared hyperspectral imaging (NIR-HSI) with machine learning, the classification model for lily of six various origins has been developed for origin traceability. Based on four algorithms, decision tree (DT), linear discriminant analysis (LDA), k-nearest neighbor (KNN) and support vector machine (SVM), all full-spectrum prediction models have an origin identification accuracy of over 90.0%. Meanwhile, the prediction models were further simplified with guaranteed accuracy by using three feature selection methods to improve the efficiency of detection in practical applications. The results show that the prediction accuracy of the model can reach 100% by the RF feature selection method. The LDA models show excellent discrimination accuracy with 100% in the origin identification. This study demonstrates the feasibility of using NIR-HSI combined with machine learning in the origin identification of lily. It provides a fast, accurate, non-destructive and comprehensive detection approach for the traceability and quality assessment of lily. In the future, the combination of NIR-HSI and algorithms promises to be a potential method for origin identification analysis, classification identification and quality evaluation in the food industry. [Display omitted] • NIR-HSI combined with machine learning to identify the origin of lily. • An accurate, rapid and non-destructive detection method for origin identification. • The LDA model demonstrates superior performance in origin identification. • Optimal feature selection model achieves 100% prediction accuracy. • The model based on repetitively selected wavelengths has an accuracy of over 90%. [ABSTRACT FROM AUTHOR]