Identification of coal categories by near-infrared spectroscopy with continuous wavelet transform-convolutional neural network
The identification of coal categories is crucial for commodity classification,quality inspection,and clean utilization.Near-infrared spectroscopy(NIRS)has gained extensive application and sustained atten-tion in various fields due to its advantages of efficiency,accuracy,environmental friendliness,and strong a-daptability.In this study,the near-infrared spectral data of 305 imported coal samples originating from 11 different countries were collected.A feature analysis of the near-infrared spectra was conducted in terms of absorbance magnitude,spectral slope,and distribution of characteristic peaks according to coal categories.The results indicated that the individual indicator could reflect the differences in coal categories,but it was insufficient for the accurate classification.A method that combined near-infrared spectroscopy with contin-uous wavelet transform-convolutional neural network(CWT-CNN)for the identification of coal categories was proposed.Through the CWT,the near-infrared spectra were transformed from two-dimensional data into three-dimensional feature images,enhancing the spectral resolution and amplifying the extraction of subtle features in spectral curves.The derived three-dimensional feature images were then input into a con-volutional neural network discrimination model based on the GoogleNet architecture.The Cross-entropy Loss function was employed as the loss function for the identification of coal categories.During the early stages of model training,the biases and weights of learning rate factor were optimized,and different optimi-zers were compared and selected.After cross-validation,the average accuracy of training set,validation set and of test set optimized model were 99.69％,96.69％and 96.39％,respectively.
NETL
NSTL
万方数据
