Machine learning enabled self-calibration single fiber endoscopic imaging

Single fiber scanners (SFSs), with the advantages of compact size, versatility, large field of view, and high resolution, have been applied in many areas. However, image distortions persistently impair the imaging quality of the SFS, although many efforts have been made to address the problem. In this Letter, we propose a simple and complete solution by combining the piezoelectric (PZT) self-induction sensor and machine learning algorithms. The PZT tube was utilized as both the actuator and the fiber position sensor. Additionally, the feedback sensor signal was interrogated by a convolution neural network to eliminate the noise. The experimental results show that the predicted fiber trajectory error was below 0.1%. Moreover, this self-calibration SFS has an excellent robustness to temperature changes (20-50°C). It is believed that the proposed solution has removed the biggest barrier for the SFS and greatly improved its performance and stability in complex environments.