Efficient and non-destructive classification of lily bulbs by laser-induced breakdown spectroscopy combined with machine learning methods

Abstract

The study presented here is a pioneering attempt to utilize Laser-Induced Breakdown Spectroscopy (LIBS) for the non-destructive testing and classification of lily bulbs. Given the complexities and time-consuming nature of traditional classification methods based on organic matter content detection, the gradual adoption of spectral technologies as an alternative tool is timely. In this research, the Isolation Forest algorithm was employed to eliminate outliers, thereby enhancing data quality. Various variable selection methods combined with five different classification algorithms were used to establish models. Both full spectrum and feature band variable classification models achieved classification accuracy rates exceeding 99% on the prediction set. Among these, the Logistic Regression (LR) model demonstrated the most exceptional performance, with classification accuracy rates of 100% in both cross-validation and prediction sets. Through the analysis of feature variable importance ranking and spectral line elements, elements such as Na, K, Fe, Ca, P, Mg, and N were identified as potentially crucial for the identification of the origin of lily bulbs. These findings demonstrate the effectiveness of using LIBS combined with machine learning to classify lily bulbs, and provide a theoretical reference for origin identification and variety recognition of other medicinal and edible foods.