Grouting asphalt concrete image segmentation based on a gated fusion mechanism

To address the challenge of segmenting multi-phase interfaces in grouting asphalt concrete (GAC), this study proposes a novel Gated Fusion Mechanism-based Dual-modal Cross-attention network (GFM-DC). The model integrates material science insight into deep learning by combining a convolutional branch for local material features and a self-attention branch for global mesostructures. Its key innovation is a gated fusion module that dynamically allocates the fusion weights of the two branch features—assigning dominant weights to the material branch in the interfacial regions to enhance boundary recognition capability, whilst emphasizing the structural branch in the internal regions to maintain regional integrity. Experiments show the model achieves an mIoU of 83.7% and an Edge Sharpness of 0.912, surpassing Swin-UNet by 5.8% and 5.4%, respectively. The Aggregate Separation Index reached 93.1%. Visual analysis confirms its precision and interpretability in complex interface segmentation.