Blended fabric with integrated neural network based on attention mechanism qualitative identification method of near infrared spectroscopy
Limei Song,Enze Chen,Tenglong Zheng,Jinyi Li,Hongyi Wang,Xinjun Zhu
DOI: https://doi.org/10.1016/j.saa.2022.121214
2022-08-05
Abstract:Near Infrared spectroscopy (NIRS) qualitative analysis technology has shown excellent development potential in the field of blend fabrics. However, the qualitative detection method based on the convolutional neural network (CNN) is difficult to accurately extract the feature of the spectral data, which will lead to missing detection or false detection; when using deep learning to build a qualitative detection model, due to interference of the external environment and other factors, the spectral data collected may have outliers, this means that the knowledge generalization on anomalous testing data, which may have a different distribution of that of the training set, is not trivial, which will also lead to missing detection or false detection. To solve the above problems, this paper proposes a novel qualitative detection neural network by analyzing the near infrared spectral data of blend fabrics. Firstly, we remove the convolutional layer and pooling layer of the CNN, making full use of the feature to enhance the feature representation ability of the model. Secondly, adding the L1 norm of the feature coefficients as a penalty term to the loss function to force those features with high redundancy to become weaker. Thirdly, in order to improve the recognition accuracy of the anomalous spectral data and minimize the model uncertainty, an ensemble machine learning approach utilizing 5 neural networks in parallel is used. To show the superiority of our proposed method, the existing methods are used as competitive methods to compare with our method. Our homemade dataset contains 3482 samples of blend fabrics with 9 different compositions. The results show that the Micro-F1-score, Micro-Specificity, Weight-F1-score, and Weight-Specificity of this method respectively 99.71%, 99.96%, 99.73%, and 99.99%, the results further confirm the method has higher analysis accuracy and stability. In addition, the method proposed in this paper can greatly improve the recognition accuracy of the anomalous spectral data. It has important practical value in the qualitative detection of blend fabrics.