Application of near-infrared spectroscopy and CNN-TCN for the identification of foreign fibers in cotton layers

Foreign fibers in cotton layers have a particular impact on the quality of the cotton. Traditional image processing methods are ineffective in detecting foreign fibers in cotton layers, which are time-consuming and costly. In order to identify foreign fibers effectively, a classification and identification method for foreign fibers in cotton layers was proposed based on NIR spectroscopy and CNN-TCN. In this study, near-infrared spectroscopy ranging from 780 nm to 2360 nm was used to identify the type of foreign fibers. Savitzky-Golay smoothing was used to preprocess spectroscopy data, and LightGBM-ANOVA was used to determine optimal wavelengths. Preprocessed spectral data extracted spectral features through the 1D convolutional neural network(1D-CNN). Then Temporal convolutional neural network (TCN), Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Recurrent Neural Network (RNN), and 1D-CNN were used to establish classification models. Compared with other time series models, CNN-TCN methods obtained better performances with the classification accuracy of over 99% in the test set and the shorter training time. The overall results illustrated that near-infrared spectral combined with the CNN-TCN method was efficient and accurate for identifying foreign fibers in the cotton layer.