Channel wave exploration methods are the preferred choice for detecting the geological conditions in coal seams. This paper intro­duces a channel wave signal feature extraction and classification method based on Local Mean Decomposition (LMD) for the identification and classification of geological structures in coal seams. Using this method, it becomes possible to detect geological anomalies hidden in the coal seam. Firstly, an in-depth investigation was conducted into the prop­agation characteristics of channel waves. A three-dimensional equiva­lent model for complex coal seams, which include goaf, scour zone, col­lapse column, and small faults, was constructed using COMSOL Mul­tiphysics. Channel wave signals were collected on this model to build a dataset. Then, a feature extraction and classification approach based on LMD and sample entropy was employed. To further optimize the fea­ture extraction process, a Refined Composite Multiscale Fuzzy Entropy (RCMFE) algorithm was integrated with LMD. Finally, classification was achieved using a Least Squares Support Vector Machine (LS-SVM) clas­sifier with a Gaussian radial basis kernel function. The research results indicate that the LS-SVM classifier used in this study achieved classifi­cation accuracies of 96.5% for the method based on LMD and sample entropy and 97.89% for the method based on LMD and RCMFE. This demonstrates that the approach proposed in this paper is suitable for the classification of small geological structures in coal mines.