Drill Core Classification Using Enhanced ResNet Model

The classification of drill core images is a critical task of geological exploration. However, the existing recognition methods for geological images struggle to accurately recognize drill core images due to subtle discrepancies between different lithological categories, complex backgrounds, unclear subjects, and lack of datasets. To address the issues, this paper constructs a drill core image dataset called Drill Core Dataset (DCD), and proposes a drill core image classification model called MSL-ResN et50, which combines self-calibration residual module and three-dimensional attention. The former is employed to capture higher-level semantic information and the latter is leveraged to mitigate interference in context information, giving higher attention to key areas. Besides, the MAM pooling strategy is utilized to alleviate the loss of features and keep consistency in feature distribution during the pooling process. The results of extensive experiments made on the DCD dataset demonstrate that MSL-ResNet50 significantly outperforms the benchmark models in drill core image classification. In particular, it attains an improvement of 4.1% in accuracy over the original model, validating its effectiveness in core image classification.