摘要
编码孔径快照光谱成像(Coded Aperture Snapshot Spectral Imaging,CASSI)技术通过单次曝光压缩成像机制,实现空间与光谱信息的高效协同获取,突破了传统光谱成像技术依赖扫描机制、数据存储传输成本高的局限。本文系统综述了CASSI技术的硬件架构、理论模型及重建算法的研究进展。硬件设计部分探讨了系统架构的迭代优化与编码孔径设计对成像性能的提升。理论模型部分分析了单色散CASSI的物理建模方法并总结了理论模型的优化路径,强调了光学误差校正对提升重建精度的重要性。重建算法部分探讨了传统重建算法的性能瓶颈,进而引出近几年基于深度学习的重建算法的突破性进展。虽然深度学习在复杂场景重建与计算效率方面展现出显著优势,但其可解释性、数据依赖性及硬件适配性仍需进一步优化。最终本文为CASSI技术展望了在系统架构、硬件革新、算法框架以及嵌入式终端开发等多维度的发展趋势,推动其在航天遥感、生物医学、深空探测及实时导航等领域广泛应用。
Coded aperture snapshot spectral imaging(CASSI)technology enables efficient synergistic acquisition of spatial and spectral information through single-shot compressive imaging.It overcomes the limitations of traditional spectral imaging techniques,which rely on scanning mechanisms and incur high costs on data storage and transmission.This paper systematically reviews the research progress in CASSI technology,focusing on its hardware architecture,theoretical models,and reconstruction algorithms.The hardware design section explores the iterative optimization of system architecture and the impact of coded aperture design on imaging performance.The theoretical model section analyzes the physical modeling methods of single dispersion CASSI and summarizes optimization paths for the theoretical models,highlighting the importance of optical error correction in improving reconstruction accuracy.The reconstruction algorithms section talks about the performance bottlenecks of traditional algorithms introducing recent breakthroughs in deep-learning based reconstruction methods.While deep learning has demonstrated significant advantages in complex scene reconstruction and computational efficiency,challenges related to interpretability,data dependency,and hardware compatibility remain to be addressed.Finally,the paper discusses the future development trends of CASSI technology across multiple dimensions,including system architecture,hardware innovations,algorithm frameworks,and embedded terminal development,aiming to promote its widespread applications in fields such as aerospace remote sensing,biomedicine,deep space exploration,and real-time navigation.
作者
朱梦琪
马杰
张培哲
尚伦艳
俞文凯
张安宁
ZHU Mengqi;MA Jie;ZHANG Peizhe;SHANG Lunyan;YU Wenkai;ZHANG Anning(Quantum Technology Research Center,School of Physics,Beijing Institute of Technology,Beijing 100081)
出处
《导航与控制》
2025年第3期202-219,共18页
Navigation and Control
基金
国家自然科学基金面上项目(编号:12474480)
国家自然科学基金重大研究计划培育项目(编号:92365115)。
关键词
计算成像
编码孔径快照光谱成像
图像重建
深度学习
优化模型
computational imaging
coded aperture snapshot spectral imaging
image reconstruction
deep learning
optimization models
