Automated and rapid target position alignment in laser–plasma experiments using deep learning algorithms

Abstract

In laser proton acceleration, laser pulses are tightly focused onto the solid target to achieve the highest intensity. For high-frequency application-oriented laser accelerators, the need for rapid and precise laser–target coupling technology is especially essential. We propose innovative methods that leverage deep learning algorithms to automate and expedite target positioning in laser–plasma experiments. Our comparative study of various techniques, such as position scanning, image recognition, and object detection, indicates that the YOLO (You Only Look Once) object detection network excels in facilitating swift and highly precise target positioning. It demonstrates an interference time of approximately 50 ms and a positioning accuracy of 8μm. Subsequently, we have successfully integrated this deep learning model into the control program of the Compact Laser Plasma Accelerator at Peking University to optimize the experimental setup process.