Gozzard David R, Wallis John S, Frost Alex M, Collier Joshua J, Maron Nicolas, Dix-Matthews Benjamin P, Vinsen Kevin
International Centre for Radio Astronomy Research, The University of Western Australia, Crawley, WA 6009, Australia.
Australian Research Council Centre of Excellence for Engineered Quantum Systems, Department of Physics, The University of Western Australia, Crawley, WA 6009, Australia.
Sensors (Basel). 2025 Sep 1;25(17):5395. doi: 10.3390/s25175395.
We present the use of a light-weight machine learning (ML) model to estimate the separation and relative brightness of two incoherent light sources below the diffraction limit. We use a multi-planar light converter (MPLC) to implement spatial mode demultiplexing (SPADE) imaging. The ML model is trained, validated, and tested on data generated experimentally in the laboratory. The ML model accurately estimates the separation of the sources to up to two orders of magnitude below the diffraction limit when the sources are of comparable brightness, and provides accurate sub-diffraction separation resolution even when the sources differ in brightness by four orders of magnitude. The present results are limited by cross talk in the MPLC and support the potential use of ML-assisted SPADE for astronomical imaging below the diffraction limit.
我们展示了使用轻量级机器学习（ML）模型来估计低于衍射极限的两个非相干光源的间距和相对亮度。我们使用多平面光转换器（MPLC）来实现空间模式解复用（SPADE）成像。该ML模型在实验室通过实验生成的数据上进行训练、验证和测试。当光源亮度相近时，该ML模型能够准确估计光源间距至低于衍射极限达两个数量级，并且即使光源亮度相差四个数量级时也能提供准确的亚衍射间距分辨率。目前的结果受到MPLC中串扰的限制，并支持ML辅助的SPADE在低于衍射极限的天文成像中的潜在应用。
