Optical tweezers with optical vortex based on deep learning

Abstract

Optical tweezers (OTs) with structured light expand degrees of freedom of particle manipulation. However, the studies of structured optical tweezers are usually accompanied by complex theoretical models, strict simulation conditions, and uncertain experimental factors, which may bring about high time costs and insufficiently precise results. In this work, we proposed a bidirectional neural network model for the analysis and design of OTs with optical vortices (OVs) as a typical structured light beam. In analyzing optical forces, the network can achieve over 98% accuracy and improve computational efficiency by more than 20 times. In further analyzing particle trajectories, the network can also achieve over 95.5% accuracy. Meanwhile, in OTs with OV-like beams, our network can still predict particle motion behavior with a high accuracy of up to 96.2%. Our network can inversely design optical vortex tweezers on demand with 95.4% accuracy. In addition, the experimental results in OTs with plasmonic vortex can be analyzed by the proposed model, which can be used to achieve arbitrary optical manipulation. Our work demonstrates that the proposed deep learning network can provide an effective algorithmic platform for the analysis and design of OTs, and is expected to promote the application of OTs in biomedicine.