<![CDATA[Seismic data interpretation is an important stage in geological exploration for identifying subsurface conditions. However, manual interpretation requires considerable time and depends strongly on expert experience. This study develops a Convolutional Neural Network (CNN) model for multi-class classification of geological interval images extracted from the Penobscot Interpretation Dataset. Seven interpreted horizons were used to divide the seismic sections into eight facies interval classes. The preprocessing stages included geological interval extraction, image resizing to 256 × 481 pixels, intensity normalization, and stratified dataset splitting into training, validation, and testing subsets. Five CNN configurations were evaluated by varying the optimizer, learning rate, hidden-layer activation function, and dropout rate. Each configuration was trained repeatedly using five fixed random seeds to evaluate performance consistency. The selected configuration used the Adam optimizer, a learning rate of 0.001, ReLU activation, and a dropout rate of 0.3. It achieved an average testing accuracy of 99.13% ± 0.17%, a macro F1-score of 0.9912 ± 0.0016, and a weighted F1-score of 0.9913 ± 0.0017. The representative run achieved 99.17% accuracy, with weighted precision, recall, and F1-score values of 0.9919, 0.9917, and 0.9917, respectively. These results indicate that the compact CNN produced high and relatively consistent performance for geological interval classification within the evaluated Penobscot dataset. External validation is still required before the model can be generalized to other seismic fields.]]>
Copyrights © 2026
