南京理工大学学报（自然科学版）2023，Vol.47Issue(6)：756-766,11.DOI:10.14177/j.cnki.32-1397n.2023.47.06.004
基于改进粒子群优化的BP神经网络图像压缩方法
Image compression method of BP neural network based on improved particle swarm optimization
摘要
Abstract
In order to improve the problem of excessive dependence on initial parameters in back propagation(BP)neural network algorithms,which leads to slow convergence speed and susceptibility to local minima,an improved particle swarm optimization(IPSO)algorithm is proposed to optimize the parameters of the BP neural network and find appropriate initial weights and thresholds.The algorithm here adds a quartile based selection strategy to the basic particle swarm optimization(PSO)algorithm,introduces the adaptive mutation probability of genetic algorithm as the disturbance probability,and adds an adaptive disturbance strategy based on the ratio of individual fitness value to population average fitness value.The IPSO-BP algorithm here significantly improves the performance of training image Lena,testing image Cameraman,and validation image Peppers.The peak signal-to-noise ratio(PSNR)and mean square error(MSE)of the IPSO-BP trained model are significantly better than particle swarm optimization with linearly decreasing inertia weights-back propagation(LDWPSO-BP),particle swarm optimization based on dynamic acceleration coefficients-back propagation(PSO-DAC-BP),and particle swarm optimization back propagation based on normal distribution(NDPSO-BP),adaptive mutation particle swarm optimization-back propagation(ADVPSO-BP),genetic algorithm-back propagation(GA-BP),and beetle antennae search-back propagation(BAS-BP),with PSNR being the largest among the seven algorithms and MSE being the smallest among them.Although the compression rate(CR)of IPSO-BP on Lena is smaller than that of PSO-DAC-BP and BAS-BP,and the CR on Cameraman is smaller than that of NDPSO-BP,ADVPSO-BP,and GA-BP,the difference is not more than 0.01 and 0.006.关键词
粒子群优化/反向传播神经网络/图像压缩/遗传算法/峰值信噪比/均方误差/压缩率Key words
particle swarm optimization/back propagation neural network/image compression/genetic algorithm/peak signal-to-noise ratio/mean square error/compression ratio分类
计算机与自动化引用本文复制引用
李敏,高岳林..基于改进粒子群优化的BP神经网络图像压缩方法[J].南京理工大学学报（自然科学版）,2023,47(6):756-766,11.基金项目
国家自然科学基金(11961001) （11961001）
宁夏自然科学基金(2022AAC02043) （2022AAC02043）
宁夏高等教育一流学科建设基金(NXYLXK2017B09) （NXYLXK2017B09）
北方民族大学重大专项(ZDZX201901) （ZDZX201901）
