<p>Lithology identification is of great significance in geological research and petroleum exploration, and its accuracy directly affects the development and utilization efficiency of oil and gas reservoirs. However, due to the diversity of igneous rock genesis, the complexity of lithological changes, and the non-uniformity of pore structure, the traditional lithology identification method is difficult to effectively deal with the complex nonlinear mapping relationships in logging data, resulting in insufficient identification accuracy and limiting the effectiveness of its practical application. To solve this problem, this study proposes a fusion model based on CNN and transformer (SE-AttenTransNet) for the Junggar Basin Hongche fault zone and uses the synthetic minority oversampling technique SMOTE to balance the dataset. The local features of logging data are extracted from the input features by CNN convolution operation, the attention of important logging response features is enhanced by using SE Block module, global modeling is performed by using transformer to capture the complex relationships in the features of long-sequence logging data, and weighted aggregation of prioritized lithology features is performed by using Attention Pooling. The experimental results show that the SE-AttenTransNet model has a significant advantage in the igneous rock lithology identification task, with a prediction accuracy of 95.15% and an AUC value of 99.60%, which significantly outperforms the traditional methods as well as a single CNN or transformer model. It shows the effectiveness and stability of the SE-AttenTransNet model in igneous rock lithology recognition, providing a novel and reliable solution for dealing with complex lithology problems.</p>