Li-Hopfield神经网络用于汉字字符识别

Li-Hopfield neural network used for Chinese character recognition

提出了一种Li-Hopfield网络,解决了Hopfield神经网络能量函数存在积分项、振荡收敛、计算复杂等问题,并继承了Hopfield网络的结构与Li网络的优点.同时发现基于部首拆分的汉字字符识别可以忽视甚至利用该网络的虚假稳定点,从而使样本存储量降为传统Hopfield网络的1.44%,表明该算法有效且快速.

In order to solve the problems in Hopfield neural networks, such as integration terms in energy function, convergence with oscillation, and complexity in computation, a novel Li-Hopfield neural network is proposed. It can not only solve the above-mentioned problems, but also inherit the structure of Hopfield neural network and the merits of Li neural network. Meanwhile, we find Chinese character recognition based on indexing components can ignore, even can utilize ,the spurious stable points（SSP） in this network. Experiments demonstrate that the sample storage for this method is reduced to 1.44% of that for the traditional method. Thus, the proposed method is effective and fast.