Multi-Feature Fusion Model Based on Attention Mechanism

In order to solve the problems of low efficiency and low accuracy of the traditional tandem feature extraction structure, this paper proposes a multi-feature fusion model based on the Attention mechanism (Attention). The model is modelled with the near-infrared spectral (NIR) dataset of marzipan and grains to verify its effectiveness in feature extraction and feature fusion. The experimental results show that the model has a significant reduction in root mean square error (RMSEP) in moisture and sugar (glucose) prediction of marzipan and grain NIR spectral data compared to the three models, namely, single CNN, LSTM and CNN-LSTM. The model shows potential in terms of accuracy and robustness in NIR data processing and prediction, and provides new ideas for NIR sensor data fusion in driving environment sensing.