Intelligent Recognition of Coal Gangue Based on Residual Network and Near Infrared Spectroscopy Technology

Abstract

This study innovatively developed a rapid classification method for coal and gangue，inte⁃ grating near infrared spectroscopy technology with a one-dimensional residual network（1D-ResNet）. To ensure the diversity of experimental samples，430 samples of coal and gangue were collected from multiple coal mines in provinces such as Henan， Hebei， and Shandong. Abnormal samples were eliminated based on Euclidean distance to obtain a high-quality dataset for modeling. Building on this， a 1D-ResNet-based classification model was constructed to accurately capture the complex mapping relationships between coal，gangue，and their spectral characteristics. This approach effec⁃ tively solved the problem of gradient vanishing and deeply mined the spectral features of coal and gangue， resulting in highly accurate analysis. After five-fold cross-validation， the model achieved an average accuracy of 96. 26%，significantly outperforming traditional machine learning algorithms such as support vector machines and random forests. The model demonstrated high consistency in the trend of loss function changes across both the training and test datasets，indicating good generaliza⁃ tion ability. Tests revealed that the model processes every hundred samples in just 16. 230 millisec⁃ onds，further highlighting its advantages and potential application value in the online sorting field for coal and gangue.