Classification of Single-Photon Emitters in Confocal Fluorescence Microscope Images by Deep Convolutional Neural Networks

Abstract

In the rapidly evolving field of quantum information technology, the accurate and efficient classification of single-photon emitters is paramount. Traditional methods, which rely on conducting time-intensive Hanbury Brown-Twiss (HBT) experiments to acquire the 2nd-order correlation function of photon statistics, are not efficient. This study presents a pioneering solution that employs Deep Convolutional Neural Networks (CNNs) to classify single-photon emitters in confocal fluorescence microscope images, thereby bypassing the need for laborious HBT experiments. Focusing on the nitrogen-vacancy centers in diamond, the model is trained using fluorescence images of emitters that have been previously classified through HBT experiments. Applied to unclassified fluorescence images, the model achieves up to 98% accuracy in classification, substantially accelerating the identification process. This advancement not only makes the classification workflow more efficient but also promises wider applicability across various color centers and isolated atomic systems that necessitate imaging for isolation verification. This research signifies a substantial advancement in the application of quantum technologies, leveraging the power of deep learning to optimize the utilization of single-photon emitters.