Rapid Focused Spot Scanning Imaging Using Multimode Fiber with a Convolutional Neural Network-Based Phase Modulation

Abstract

In this paper, we present a rapid beam-focusing method for multimode fiber (MMF) that integrates a Convolutional Neural Network (CNN) with a Spatial Light Modulator (SLM). This approach efficiently focuses light fields by training neural networks to approximate the transmission matrix of the optical system, eliminating the need for iterative calculations. Experimental results indicate that this method can generate focused spots within 200 ms, representing a significant improvement over traditional methods. Additionally, the feasibility of point-scanning imaging was verified using the USAF-1951 test target with a resolution greater than 7.81 μm. This advance is expected to improve the speed and accuracy of endoscopic imaging and other real-time imaging applications, potentially leading to more effective diagnostic tools and imaging technologies in medical and industrial settings.

Keywords

• multimode fiber
• beam focusing
• deep learning
• scanning imaging
• wavefront shaping