Laser-Induced Breakdown Spectroscopy–Visible and Near-Infrared Spectroscopy Fusion Based on Deep Learning Network for Identification of Adulterated Polygonati Rhizoma

Abstract

The geographical origin of foods greatly influences their quality and price, leading to adulteration between high-priced and low-priced regions in the market. The rapid detection of such adulteration is crucial for food safety and fair competition. To detect the adulteration of Polygonati Rhizoma from different regions, we proposed LIBS-VNIR fusion based on the deep learning network (LVDLNet), which combines laser-induced breakdown spectroscopy (LIBS) containing element information with visible and near-infrared spectroscopy (VNIR) containing molecular information. The LVDLNet model achieved accuracy of 98.75%, macro-F measure of 98.50%, macro-precision of 98.78%, and macro-recall of 98.75%. The model, which increased these metrics from about 87% for LIBS and about 93% for VNIR to more than 98%, significantly improved the identification ability. Furthermore, tests on different adulterated source samples confirmed the model’s robustness, with all metrics improving from about 87% for LIBS and 86% for VNIR to above 96%. Compared to conventional machine learning algorithms, LVDLNet also demonstrated its superior performance. The results indicated that the LVDLNet model can effectively integrate element information and molecular information to identify the adulterated Polygonati Rhizoma. This work shows that the scheme is a potent tool for food identification applications.

Keywords

• LIBS
• VNIR
• Polygonati Rhizoma
• deep learning
• adulteration