关键词: 深度置信网络（DBN）, 量子粒子群算法（QPSO）, 形状检测, 深度学习

Abstract:

In order to improve the convergence accuracy of the standard Quantum Particle Swarm Optimization（QPSO） algorithm, a calculation method of the average optimal position weight of particles based on the change rate of the length of the particle potential well is proposed. By adjusting the average optimal position, the optimization ability of particles is improved. The improved QPSO algorithm is applied to the optimization of the learning rate parameters of the Depth Belief Network（DBN） model, so as to find the optimal DBN model parameters to improve the recognition accuracy of the DBN model. Finally, the DBN network （LQ_DBN）, which is optimized by the improved QPSO algorithm, is applied to the egg yolk shape detection in the experiment. Compared with the existing typical DBN network model, the recognition accuracy of LQ_DBN model in egg yolk shape detection experiment is higher than that of CC-PSO-DBN, PSO_MDBN and standard DBN models, and the stability of recognition accuracy is the highest among the four comparison models, which shows that the DBN network model based on the improved QPSO algorithm has achieved better optimization effect.

Key words: Depth Belief Network（DBN）, Quantum Particle Swarm Optimization（QPSO）, shape detection, deep learning