A machine learning assisted approach to classify rose species and varieties with laser induced breakdown spectroscopy.

Background: Classification of rose species and verities is a challenging task. Rose is used worldwide for various applications, including but not restricted to skincare, medicine, cosmetics, and fragrance. This study explores the potential of Laser-Induced Breakdown Spectroscopy (LIBS) for species and variety classification of rose flowers, leveraging its advantages such as minimal sample preparation, real-time analysis, and remote sensing. A Q-switched Nd: YAG laser, operating at 532 nm with an optimal energy of 290 mJ and a pulse duration of 5 ns, was used to generate plasma on the surface of rose samples.

Results: The study demonstrated that laser induced breakdown spectroscopy effectively provided elemental analysis of different rose species and varieties. However, LIBS alone was insufficient for an accurate classification, necessitating the integration of machine learning techniques. Principal component analysis (PCA), an unsupervised model, was used to reduce dimensions of large datasets, however, it failed to provide precise classification results. To enhance accuracy, various supervised machine learning models were employed. The Quadratic Support vector machine (SVM) achieved a test accuracy of 95 % for classifying rose petals requiring a computational time 101.59 s. In contrast, Linear Discriminant Analysis (LDA) attained 100 % test accuracy for leaves with minimal computational time 50.59 s. These findings highlight that machine learning assisted LIBS significantly improves classification accuracy, especially when spectral similarities complicate differentiation.

Significance And Novelty: To the best of our knowledge, this study is the first to utilize LIBS for rapid classification of rose species and varieties. This approach offers a powerful, efficient and reliable method for identifying rose varieties and species based on spectral data, making it valuable for botanical and agricultural applications. Furthermore, machine learning-assisted LIBS can be explored in the future to classify samples where variations exist only in intensity while the elemental composition remains the same.