Generation of the Flat-top beam using convolutional neural networks and Gerchberg-Saxton algorithm

Abstract Laser technology has made rapid progress in recent years and has been widely used in various fields such as medicine, biology, military, and materials science. However, the limitations of traditional Gaussian intensity distribution of the laser beams in applications have prompted the emergence and development of flat-top beam shaping technology, which has received widespread attention. Here, we introduce a new method for generating flat-top beams that combines the traditional Gerchberg-Saxton algorithm with convolutional neural networks, using spatial light modulators to achieve flat-top beam shaping. A comparative analysis was conducted by comparing the root mean square error and diffraction efficiency of the generated flat-top beam with the results obtained using only the traditional Gerchberg-Saxton algorithm. Compared with the traditional Gerchberg-Saxton algorithm, the method proposed in this paper can generate a flat-top beam with smaller differences from the target light intensity and higher energy utilization, providing new possibilities for the application of laser technology.