Fast-convergent multiple-plane phase retrieval using evolutionary crossover strategies

Abstract

Insufficient intensity variation in iterative phase retrieval algorithms can lead to stagnation and slow convergence due to limited measurement diversity. This necessitates robust reconstruction methods capable of recovering the phase from intensity measurements by escaping the local minima. In this paper, evolutionary crossover strategies, specifically single-point, two-point, and uniform crossover, are investigated for their effectiveness in improving phase reconstruction and minimizing the amplitude MSE within the Single Beam Multiple Intensity Reconstruction (SBMIR) framework. Results show that uniform crossover achieves fastest convergence and lower reconstruction error compared to the single-point and two-point crossover methods. At later iterations, it attains the lowest amplitude MSE but exhibits higher variability (s = 0.079) relative to single- and two-point crossover due to its efficient exploration of the solution space. These findings highlight a trade-off between convergence speed and stability, providing valuable insight into the selection of crossover strategies for phase retrieval applications.