Infrared Image Semantic Segmentation Based on Text Feature Fusion

This paper discusses the application of semantic segmentation model fused with text features in the field of infrared image segmentation. The proposed method uses the text encoder of the CLIP to extract semantic information according to the text label and fuse it into the semantic segmentation model. In addition, two innovative modules are designed: the Mul-scale aggregation module (MAM) and the Cross-modal fusion module (CFM). Among them, the MAM module uses deep dilated convolution at different scales to obtain multi-scale features and effectively fuse them to enhance the model's perception ability of multi-scale targets. The CFM module adopts the cross-attention mechanism to efficiently fuse text features and image features, making full use of the complementary information of the two modalities. Sufficient experiments have proved the effectiveness of this method in significantly improving the performance of infrared image semantic segmentation models.