Improved 1D convolutional neural network adapted to near-infrared spectroscopy for rapid discrimination of Anoectochilus roxburghii and its counterfeits

Anoectochilus roxburghii (Wall.) Lindl. (Orchidaceae) is a rare traditional Chinese medicine. For seeking high profit, some traditional Chinese medicine sellers usually adulterated A. roxburghii with Goodyera Schlechtendaliana and Ludisia discolor or directly fake A. roxburghii using Anoectochilus formosanus. These counterfeits with similar appearance greatly influence the prescription efficacy. Therefore, there is an urgent need for an effective and fast authentication method to identify A. roxburghii and its counterfeits. In this paper, the near-infrared spectroscopy (NIRS) data of A. roxburghii and its counterfeits are mearsured. Then, an improved inception architecture based 1-dimensional convolutional neural network (Improved 1D-Inception-CNN) is designed for processing the NIRS data and identifying A. roxburghii and its counterfeits. The Improved 1D-Inception-CNN has less parameters and high calculation efficiency which makes the identification model more practical. The experimental results show that compared with traditional structured CNN models, the complexity of the Improved 1D-Inception-CNN is reduced by 40 %, the parameters are reduced by 50 % and the performances are improved by 1.01 %. Therefore, the Improved 1D-Inception-CNN model based on NIRS technology can effectively and quickly identify A. roxburghii and its counterfeits. (c) 2021 Elsevier B.V. All reserved.

Automated summary

This study employs near-infrared spectroscopy data to develop an improved neural network model for identifying Anoectochilus roxburghii and its counterfeits. The improved model simplifies the structure, enhances performance, and is able to effectively and quickly distinguish between genuine and fake A. roxburghii.