Development of analytical method associating near-infrared spectroscopy with one-dimensional convolution neural network: a case study

Near-infrared spectroscopy (NIRS) is an outstanding detection technology with fast, nondestructive, on-line characteristics, which has been used in many fields to achieve qualitative/quantitative detection of products. This study proposes an analysis method that combines a one-dimensional convolution neural network (1-CNN) with NIRS to select a characteristic wavelength for a subsequent qualitative/quantitative analysis, while considering fully dimensional NIRS information. Batch normalization (BN) and an exponential decay learning rate are embedded into the training stage of the 1-CNN which improves performance and reduces the risk of overfitting. The proposed method is used for NIRS of dried matsutake samples from four regions in China, and a geographical origin identification model is established for the matsutake. The developed 1-CNN model achieves better classification accuracy than traditional models. The experimental results demonstrate that the proposed 1-CNN achieves good generalization and can act as an alternative method for analyzing NIRS data.