摘要
提出了一种基于随机退火机制的竞争层神经网络学习算法,并将其应用于解决图像特征绑定问题。该算法将竞争层神经网络的串行迭代模式改为随机优化模式,通过采用退火技术避免网络收敛到能量函数的局部极小点。通过理论分析证明了该算法与竞争层神经网络动力学方程的等价性。通过对比实验验证了算法能够在加快网络收敛速度的同时提高特征绑定结果的合理性。
A competitive layered neural network learning algorithm based on random annealing is proposed and applied in solving image feature binding problem.The proposed algorithm,instead of using serial iteration,uses a random optimization method in learning process;prevents the network from trapping into local optimum through using of the annealing technique.Theoretical analysis proves that the proposed algorithm is equivalent to the dynamics of the competitive layered neural network.Comparative experiments show that the proposed algorithm is capable of speeding up the network convergence as well as improving the rationality of the results of feature binding.
出处
《计算机工程与应用》
CSCD
北大核心
2010年第19期39-42,47,共5页
Computer Engineering and Applications
关键词
竞争层神经网络(CLMNN)
神经网络
图像处理
特征绑定
Competitive Layered Model Neural Networks（CLMNN）
neural networks
image processing
feature binding
