A context-aware multi-scale pyramid-based network for three-dimensional palatal rugae cross-domain recognition

Considering that the feature information of palatal rugae cross-sectional images varies under different acquisition methods, while exhibiting certain consistency in detailed texture features and positional information across different data domains, this paper proposes a context-aware multi-scale feature pyramid fusion network for three-dimensional palatal rugae cross-domain recognition. First, a learnable Gabor filter is designed to extract the texture features of three-dimensional palatal rugae, leveraging the advantage of deep learning frameworks that can adaptively learn and optimize filter parameters, allowing the Gabor filter parameters to no longer be restricted to fixed preset values. Secondly, a contextual attention mechanism is introduced to enhance the sensitivity of the recognition network to the global features of palatal rugae, thereby ensuring the network's adaptability to different palatal rugae data domains. Finally, a progressive feature fusion strategy is employed to integrate the features output by the multi-scale feature pyramid, which helps provide a more comprehensive and rich feature representation for palatal rugae. Experimental results indicate that this method significantly improves the accuracy of palatal rugae recognition on cross-domain datasets.