计算机科学
人工智能
生成语法
桥接(联网)
人工神经网络
数据驱动
判别式
数据采集
自组织
深度学习
机器学习
强化学习
光学现象
稳健性(进化)
人工生命
无监督学习
计算智能
生成模型
人工智能应用
作者
Jin Zhou,Junyang An,Junzhuan Wang,Yibo Xu,Xiaomu Wang
摘要
Scanning near-field optical microscopy (SNOM) stands at the forefront of nanoscale optical characterization, yet faces inherent challenges including low signal-to-noise ratio, complex physical interpretation, and slow data acquisition. This review comprehensively explores the paradigm shift driven by artificial intelligence (AI), especially machine learning, in overcoming these limitations. We detail how discriminative models enable robust image enhancement and super-resolution reconstruction, while generative models such as generative adversarial networks and diffusion models push the boundaries of perceptual realism. We highlight the critical role of AI in quantitative nano-spectroscopy, where physics-informed neural networks and hybrid models directly invert near-field signals to extract intrinsic material properties like dielectric functions and polaritonic parameters with unprecedented speed and accuracy. Furthermore, we examine the emergence of intelligent, autonomous data acquisition systems, where reinforcement learning and compressed sensing strategies dramatically accelerate experimental measurement and optimization processes. However, the integration of AI into SNOM presents distinctive challenges, such as ensuring data fidelity, managing system complexity, and bridging the simulation-to-reality gap. This review argues that the path to trustworthy AI in SNOM hinges on a synergistic framework that tightly integrates rigorous physical models, explainable AI architectures, and robust validation protocols. By synthesizing these advancements, we envision the evolution of SNOM into a high-throughput, quantitative nano-analytical platform, fundamentally transforming its capacity for discovery in materials science and nanophotonics.
科研通智能强力驱动
Strongly Powered by AbleSci AI