Abstract:

In recent years， the low-altitude economy has emerged as one of the focuses in both theoretical research and engineering applications. Among the various theoretical and technical approaches underpinning the low-altitude economy， information fusion plays a pivotal role. However， in complicated environments， it still encounters formidable challenges， including significant heterogeneity among information sources， lack of spatial-temporal consistency， data sparsity and dirty， and difficulties in cooperative interoperability. With the rapid development of deep learning， the theories and methods for multi-source information fusion have undergone significant advancements in recent years. These developments have enabled information fusion to become more intelligent， end-to-end， robust， and highly adaptive， which aligns well with the urgent needs of applications within the low-altitude economy. This paper discusses the evolution of multi-source information fusion theory in the era of deep multi-modal learning and its applications in the low-altitude economy， with a particular focus on its advantages， opportunities， and， especially， the remaining challenges. Furthermore， the directions and pathways for the research of information fusion theories and methodologies in the context of the low-altitude economy are explored.

Key words: low-altitude economy, deep learning, multi-source information fusion, deep fusion, multi-modal learning