A study on multisource sensory data fusion methods in GIS devices based on multimodal deep learning

This paper studies a multi-source sensing data fusion method for GIS devices based on multimodal deep learning. Across-modal Transformer architecture fusion model is proposed, which effectively integrates heterogeneous data such as images, sensor time series, and geographic locations through data preprocessing, modality-specific encoders, and cross-attention mechanisms. The model introduces a composite loss function and the AdamW optimization strategy to improve fusion accuracy and robustness. Experimental results show that the proposed model significantly outperforms traditional baseline methods in terms of accuracy, root mean square error, and F1 score, demonstrating the advantages of cross-modal attention in capturing complementary information between modalities. This method provides an efficient and reliable technical solution for GIS device sensing applications and has significant engineering value.