数据驱动的多源遥感信息融合研究进展

Abstract

多源遥感信息融合技术是突破单一传感器的观测局限, 实现多平台多模态观测信息互补利用, 生成大场景高“时-空-谱”无缝的观测数据的重要手段。随着人工智能理论与技术的日益完善, 数据驱动的多源遥感信息融合获得了研究者的广泛青睐, 然而, 数据驱动算法与生俱来的低物理可解释性, 弱泛化能力都阻碍了其在多源遥感信息融合领域的长远发展。因此, 本文分别对同质遥感数据融合, 异质遥感数据融合, 以及点-面融合的有关研究成果进行了系统的梳理和归纳, 分析了各融合问题的发展趋势。最后, 对算法研究进展进行了总结, 剖析了数据驱动的融合算法所面临的挑战, 指出了未来多源遥感信息融合领域的研究方向。

Alternate abstract:

Multi-source remote sensing data fusion is an important technology to generate seamless observation data of large scene with a high temporal-spatial-spectral resolution, which breaks through the limitation of single sensor observation and realize the complementary utilization of multi-platform and multi-mode observation data. With the improvement of artificial intelligence theory and technology, data-driven multi-source remote sensing data fusion has been widely favored by researchers. However, the inherent low physical interpretability and weak generalization ability of data-driven algorithms have impeded its further development in multi-source remote sensing data fusion. Therefore, this paper systematically summarizes the researches of homogeneous remote sensing data fusion, heterogeneous remote sensing data fusion and point-surface fusion through three sections, and analyzes the trend of each fusion problem. Finally, this paper discusses the challenges faced by data-driven fusion algorithm, and points out some feasible future directions of multi-source remote sensing data fusion, which provides some suggestions for researchers in this field.