An improved differential evolution algorithm based on basis vector type and its application in fringe projection 3D imaging

According to the different generation methods, the types of candidate position (basis vector) in the meta-heuristic algorithm can be divided into stochastic, deterministic, and probabilistic. The stochastic type is beneficial to the preservation of population diversity, deterministic model can accelerate convergence, and probabilistic style emphasizes the balance between population diversity and convergence speed. In order to make full use of various characteristics, an improved differential evolution algorithm based on basis vector type (DEBVT) is proposed. On the one hand, in the process of mutation, individuals in the population choose the appropriate basis vector form for evolution in light of their features and needs. On the other hand, the control parameters are adaptively adjusted for the purpose of balancing exploration and exploitation. Based on twenty-nine benchmark functions, DEBVT is compared with several meta-heuristics and differential evolution variants with different types of basis vectors. All experimental results demonstrate that compared with other competitive algorithms, the optimization performance of the proposed DEBVT is remarkable. In addition, an improved UNET network (IUnet) is proposed for absolute phase extraction in fringe projection 3D imaging technology, and the activation function configuration of the network is optimized by DEBVT to construct the IUnet-DEBVT network, which further reduces the error of absolute phase and verifies of the effectiveness of DEBVT. (c) 2023 Elsevier B.V. All rights reserved.