Abstract: Faults play a crucial role in hydrocarbon migration and accumulation, making the efficient and accurate identification of seismic faults essential for oil and gas exploration and development. Traditional manual interpretation methods are inefficient and highly subjective, failing to meet practical demands. Although deep learning has been successfully applied to fault detection and has significantly improved the efficiency of seismic data interpretation, challenges remain in accurately delineating faults under complex geological conditions. To address these challenges, AmpAnt-Net is proposed, which integrates a Convolutional Neural Network (CNN) and a Graph Convolutional Network (GCN) to achieve dual-attribute parallel feature extraction, while introducing a cross-attention machanism in the bottleneck layer to realize cross-attribute feature fusion. In the encoder, the model utilizes CNN modules to extract amplitude attribute features, capturing texture and reflector continuity, while employing GCN modules to model the spatial connectivity and orientation consistency of ant-tracking attributes. At the bottleneck, the cross-attention module achieves effective feature fusion and enhancement across the two modalities, and in the decoder, multi-scale features are progressively integrated to reconstruct fault structures. Experimental results demonstrate that AmpAnt-Net outperforms UNet and UNet++ across multiple evaluation metrics on both synthetic and real seismic datasets, showing significant advantages in detecting small-scale faults and characterizing complex fault zones, and exhibiting strong generalization ability and application potential.