BP Neural Network Image Restoration Based On Intelligent Optimization Algorithm

With the development of society,video and images are the main ways for people to obtain information.Due to the influence of factors such as motion,electronic interference,atmospheric turbulence,etc,the image quality is reduced.The purpose of image restoration is to make the restored image closer to the original image.Starting from the degradation model,this paper uses traditional image restoration techniques,such as Wiener filtering,L-R restoration,blind deconvolution restoration,etc,and conducts image restoration experiments.It is found that the restoration effect is not ideal,and traditional methods require high prior knowledge,but often a lot of prior knowledge cannot be obtained.The BP neural network has the advantages of autonomous learning and strong generalization ability.It uses BP neural network to perform image restoration,so that it can get rid of the dependence on prior knowledge.The experiments shows that the BP neural network algorithm has better image restoration effects than traditional methods,but the BP neural network is sensitive to the initial weight threshold,the convergence speed is slow,and it is easy to fall into the local optimal solutions.In order to improve the BP neural network's sensitivity to initial weights and thresholds,and easy to fall into local optima,this paper establishes a cuckoo algorithm to optimize the BP neural network image restoration model.The cuckoo algorithm is used to search the initial weights and thresholds of the BP neural network,eliminating the dependence of the BP neural network on the initial weights and thresholds,improve the step length and discovery probability of the Cuckoo algorithm,and the superiority of the improved algorithm is verified through typical test function.Use the improved cuckoo algorithm to search the initial weight threshold of the BP neural network,and restore the Lena image and Cameraman image with the optimized network.The experimental results show that the ICS-BP neural network is better than the BP neural network and the PSO-BP neural network for the Lena image.The peak signal-to-noise ratio(PSNR)of the network has increased by 4.83% and 3.53%,and the structural similarity(SSIM)has increased by 1.62% and 0.67%,which shows that the optimized BP neural network has a better image restoration effect.At the same time,ICS-BP converges in 35 iterations,which is faster than the 137 convergence of the BP neural network and the 81 convergence of the PSO-BP neural network,thus improving efficiency.In order to improve the convergence speed of the BP neural network,a BP neural network algorithm based on the sparrow search algorithm is established,In view of the shortcomings of the sparrow algorithm,the population diversity is reduced in the later stage,and it is easy to fall into the local optimum.Combine the sparrow search algorithm with Levy flight and add the weighted inertia factor,and conduct simulation experiments on ISSA and SSA through seven test functions.The experimental results show that ISSA is superior to SSA in terms of solution accuracy and convergence speed.The improved sparrow search algorithm is used to optimize the initial weights and thresholds of the BP neural network and apply them to image restoration.The experimental results show that the ISSA-BP network model converges after 22 image restoration iterations,which is faster than ICS-BP's 35 Convergence times,81 times of convergence of PSO-BP,and 137 times of convergence of BP neural network,improve the convergence speed.For Lena images,compared with BP neural network and PSO-BP neural network,the peak signal-to-noise ratio(PSNR)of ISSA-BP neural network increased by 1.37% and 0.11%,and the structural similarity(SSIM)increased by 0.99% and 0.04%,which shows that the ISSA-BP neural network image restoration effect is better and the restored image is clearer.In general,the peak signal-to-noise ratio and structural similarity of the image restored by ICS-BP neural network are the largest,the restored image is closer to the original image,and the number of iterations is reduced,which improves the efficiency and can jump out of the local optimal solution.