Optimization Algorithm And Application Based On Fruit Fly Visual Response Mechanism And Population Evolutio

Flies as an insect are extremely sensitive to the changes of a moving target,and meanwhile their visual systems can perceive the movement behaviors of targets in real-time environments.Such a unique visual response feature provides a new bio-theoretical basis for computer vision.This will help to explore an interdisciplinary branch between computer vision and evolutionary computation after combining the fly’s visual information processing mechanism with the mechanism of population evolution.Therefore,with the help of the visual neurobiological theory and species evolution mechanisms,the thesis develops three kinds of visual evolutionary neural models and related algorithms to solve single-objective optimization problems with inequality or equation constraints and two-and many-objective optimization problems with bounded variable constraints,respectively.Hereafter the theoretical and experimental analyses are discussed about their computational complexity and performances.The acquired results have a certain role in promoting the fusion of evolutionary computation and computer vision,and can provide new solutions for optimization problems.The main tasks and achievements are summarized below:A.In order to solve the problem of single-objective function optimization with equality and inequality constraints,a state matrix is constructed by taking any candidate solutions as a state,and meanwhile the objective function value of the state is regarded as a gray value.Then,related to the theory of the fly’s visual neurophysiologic findings,an improved fly visual neural network model is established to output a learning rate after each image is inputted.Further,in the process of solution search,each state matrix is updated in terms of an improved strategy of population evolution.These forms a fly visual evolutionary neural network.The theoretical analysis on computational complexity shows that the network’s computational complexity is mainly determined by its input resolution and the dimensions of optimization problems themselves.Numerical experiments have validated that the model’s algorithm is effective and stable.B.For the problem of two-objective optimization with bounded variable constraints,an improved fly visual neural network with simple structure and few parameters is developed to output the learning rate after an image frame is inputted.Thereafter,an improved particle update strategy is designed to transform each state matrix into a new matrix with the help of the conventional particle position update rule appearing in the basic particle swarm optimization approach.Then,the visual neural network is integrated with the update strategy to generate a fly visual evolutionary neural network,in which an external file set is developed to store high-quality solutions and update them by the classic non-dominated sorting method and crowding distance approach.Comparative experiments show that the visual evolutionary neural network can achieve effective solution search stably.C.Aiming at the problem of many-objective optimization with bounded variables,an improved fly visual neural network,which can generate multiple kinds of learning rates,is designed by improving and simplifying the structure of the above fly visual neural network.Then,a multistrategy state update rule is used to transfer each state matrix into a new matrix in the process of state transition.The external archive is updated by a relaxed dominance-based non-dominated sorting method and an angle selection strategy-based crowding distance method.These are combined to form a fly visual evolutionary neural network.Comparative experiments verify that the network’s algorithm is clearly superior to the compared approaches with the aspects of solution quality and solution search stability.