Prediction of multispectral staining recipe for Mongolian scots pine based on convolutional neural network–bidirectional gated recurrent unit (CNN–BiGRU) with attention mechanism network model

Abstract Fast-growing wood that imitates colours develops colour differences over time. This shortens its service life. To solve the problems of short colour retention time and insufficient accuracy of colour imitation. In this study, a prediction method for Mongolian scots pine dyeing formulation based on 1D convolutional neural networks (CNN), Bidirectional gated recirculating unit network, linear discriminant analysis and attention mechanism is proposed. First, a large database was established through experiments involving the dyeing of Mongolian scots pine wood chips and multispectral measurements. Then, linear discriminant analysis was used for feature extraction, classification and dimension reduction of multispectral information to reduce the size of data input. Then, the data were fed into a one-dimensional convolutional neural network for feature re-extraction and recipe prediction by a two-way gated recurrent unit network, while an attention mechanism was introduced to highlight key spectral segments and improve the efficiency of the network model. The evaluation results show that the segments and improve the efficiency of the network model. The evaluation results show that the multispectral data input significantly improves the colour difference problem of colour imitation over time, and the present model significantly improves the accuracy of colour imitation. According to the International Commission on Illumination (CIE) 2000 color difference formula (CIEDE2000), the model achieved a minor-difference grade of 99.39% and a no-difference grade of 85.63%, with a coefficient of determination ( R 2 ) of 0.95.