Recognition Method of Coal-Rock Reflection Spectrum Using Wavelet Scattering Transform and Bidirectional Long-Short-Term Memory

Classifying and recognizing the reflection spectrum of coal-rock is an innovative method for coal-rock identification in coal mining process. Herein, a classification and recognition method of coal-rock reflection spectrum based on wavelet scattering transform (WST) and bidirectional long-short-term memory (BiLSTM) network was proposed to improve the recognition speed and accuracy. First, the reflection spectra of coal-rock samples were obtained using the coal-rock reflection spectrum information acquisition platform, and two spectral databases with different coal-rock states and different sampling parameter combinations were established to train the network model. Second, the original data were preprocessed by Gaussian filtering and randomly divided into the training set and test set. The wavelet scattering network was used to effectively extract spectral features from the reflection spectrum and generate a feature matrix. Finally, the training set feature matrix was input into the BiLSTM network model for training to obtain the WST-BiLSTM model. The effectiveness of the proposed network model was verified using the test set. The experimental results showed that the WST-BiLSTM model can classify and identify the coal-rock reflection spectrum more accurately than other related models in literature, and the recognition accuracy for the two databases reached 99.4% and 100%. Based on the constructed multi-state and multi-parameter combination spectral database, the proposed coal-rock recognition model has good adaptability to the reflected spectrum collected by different parameters. Hence, this model can provide a theoretical basis and technical premise for automatic and intelligent coal mining. A reflection spectrum database is established with different coal-rock states and sampling parametersA coal-rock reflection spectrum recognition model is developed using wavelet scattering transform feature extraction method.Training speed and recognition accuracy of the model are improved by changing the sampling parameters.