Coal and gangue classification in actual environment of mines based on deep learning

Hengxuan Luan,Hengxuan Luan,Hao Xu,Wei Tang,Ying Tian,Qiang Zhang,Hengxuan Luan,Hao Xu,Wei Tang,Ying Tian,Qiang Zhang

DOI: https://doi.org/10.1016/j.measurement.2023.112651

IF: 5.6

2023-04-01

Measurement

Abstract:Existing deep learning technologies used to classify coal and gangue are mostly carried out in laboratory environment, which cannot meet the needs of actual coal mining process. To apply in actual environment, we propose an image window pixel information interaction method that synergizes the strengths of Convolutional Neural Networks (CNN) and Swin Transformer. Specifically, we propose a coal and gangue classification framework which combines traditional convolutional neural networks for feature extraction of coal and gangue, while optimizing the model using Batch Normalization (BN) layers and freezing fully connected layer (FFCL) weights. We evaluate our method using real data gathered from coal mines and compare results to some existing methods, demonstrating that our method can classify coal and gangue with high accuracy in actual environment, achieving approximately 95% in precision, recall, accuracy, and F1 Score. Our method paves the way for accurately classifying coal and gangue in actual environment of mines.

engineering, multidisciplinary,instruments & instrumentation