Terahertz coal-rock identification method based on multi-representation feature fusion

To improve the classification performance of coal-rock identification by enhancing spectral feature representation, this study proposes a terahertz-based multi-representation feature fusion model, termed THz-CVFusionNet. Terahertz time-domain signals of six types of coal-rock samples were acquired using a terahertz time-domain spectroscopy (THz-TDS) system. After frequency-domain transformation, the 0.1-2.0 THz range was selected for spectral analysis, and the effective absorption coefficient was extracted as the key feature parameter. A dual-branch network was developed to integrate complementary spectral representations. The 1D-CNN branch captures local spectral patterns from effective absorption coefficient sequences. The spectral sequences were further transformed into two-dimensional GAF representations for global correlation modeling using a 2D Vision Transformer. Feature-level fusion was then used to combine the two branches for classification. Experimental results show that the proposed model achieved accuracy, macro-precision, macro-recall, and macro-F1 score values above 97%, outperforming several mainstream models. These results demonstrate the potential of THz-TDS combined with feature fusion for accurate coal-rock identification.