断层的检测和识别在煤炭勘探开采过程中至关重要，传统的人工解释断层方法已经无法满足实际生产的需求，基于深度学习的地震断层解释方法在断层分割领域表现较为出色。常规卷积神经网络(CNN)感受野有限，不能很好地利用全局信息，会导致一些预测的断层存在连续性不足和断层缺失等问题。Transformer具有提取全局信息的优势，引入CNN和Transformer融合的TransUNet网络，构建一种基于CBAM-TransUNet的地震断层识别方法对二维地震断层图像进行识别。首先，将CBAM-Block注意力模块融入TransUNet网络，将该模块分别加入CNN断层编码器部分和连接断层编码器与断层解码器的3层跳跃连接部分，同时从通道和空间2个维度增强地震断层图像的识别能力；其次，选择Dice损失函数和交叉熵损失函数联合优化的损失函数，使得断层图像分割更为准确，CBAM-TransUNet断层识别网络在合成地震数据集上获得的DICE值和IOU值分别提高到0.84和0.75，试验结果表明断层识别的连续性更强，明显优于其他经典分割方法；最后，利用构建的模型对荷兰近海北海F3区块真实地震数据集进行了断层解释。试验结果表明：基于CBAM-TransUNet的地震断层识别方法在去除冗余断层信息的同时能够有效识别出断层，在断层识别准确度和断层识别连续性方面表现优异，识别出的断层细节更加丰富，提高了断层识别的精度，可以有效应用于实际地震数据中识别断层。

Detection and identification of faults are crucial in the process of coal exploration and mining, and the traditional manual method of fault interpretation can no longer meet the needs of actual production, and the deep learning-based seismic fault interpretation method performs better in the field of fault segmentation. Conventional convolutional neural network (CNN) has limited sensory field and cannot make good use of the global information, which will lead to some predicted faults with insufficient continuity and missing faults, etc. Transformer has the advantage of extracting global information, and introduces the TransUNet network which is a fusion of CNN and Transformer to construct a CBAM- based seismic fault identification method. TransUNet seismic fault identification method to identify 2D seismic fault images. Firstly, the CBAM-Block attention module is integrated into the TransUNet network, and the module is added into the CNN tomography encoder part and the 3-layer jump connection part connecting the tomography encoder and the tomography decoder, respectively, to enhance the recognition ability of the seismic tomography image from two dimensions, namely, the channel and the space; secondly, the loss function optimised jointly by the Dice loss function and the cross-entropy loss function is selected to make the segmentation of the tomography image more accurate. function and cross-entropy loss function to make the fault image segmentation more accurate, and the DICE and IOU values obtained by the CBAM-TransUNet fault identification network on the synthetic seismic dataset are increased to 0.84 and 0.75, respectively, and the experimental results show that the continuity of the fault identification is stronger, which is obviously superior to other classical segmentation methods; finally, the constructed model is used to interpret the faults on the real seismic dataset of the F3 block of the North Sea, off the coast of the Netherlands. Finally, the constructed model was used to interpret the faults in the real seismic data set of Block F3 in the North Sea off the Netherlands. The experimental results show that the seismic fault identification method based on CBAM-TransUNet can effectively identify the faults while removing the redundant fault information, and performs well in terms of fault identification accuracy and fault identification continuity, and the identified faults are richer in details, which improves the accuracy of fault identification, and can be effectively applied to identify the faults in the real seismic data.