Top Coal Caving Control Method Based on Image Segmentation

XIONG Wu

doi:10.16545/j.cnki.cmet.2025.06.007

Abstract：To enhance the intelligence level of manual coal caving,a caving control method based on image segmentation was proposed.By analyzing the difference between coal blocks and gangue under different illuminations,the illumination value with the maximum difference was selected for image acquisition.Combined with the actual underground production conditions,a dust removal algorithm based on dark channel was proposed.A semantic segmentation model with dual task heads for gangue and foreground was designed to classify pixels in the image into foreground and gangue.After obtaining the classification results,the mixed gangue rate was calculated by calculating the proportion of the gangue region in the foreground region,and a closing threshold was set to achieve intelligent control of the coal caving process.The trained segmentation model was deployed on the 81202 working face,and the segmentation accuracy of the model was verified to meet the actual production requirements through testing the segmentation results.With the mixed gangue rate closing threshold set at 20％,the ash analyzer was used to detect the coal quality of the 8 cuts after deploying the model.The results showed that the mixed gangue rate was 17.6％,proving that the proposed control method could meet the production control accuracy requirements.