用于手写体数字识别的视觉与神经网络混合算法

本文提出了用于无约束手写体数字特征提取和识别的视觉与神经网络混合算法。为了提高不变性特征的稳定性及网络收敛速度，我们引入周期包括函数来取代传统的sigmoid激活函数，计算机模拟结果显示该算法及激活函数能有效地提取手写体不变性特征，提高网络收敛速度和识别率。

A Hybrid Algorithm of Vision Learning and Neural Networks for Recognition of Handwritten Digits

A learning algortithm for invariance extraction and recognition of unconstrained handwritten digits is proposed in the article.Furthermore,a novel periodic packet activation funnction is suggested to replace the traditional sigmoid activation function to reduce the sensitivity of the extracted features to samples with large variance and to improve the learning speed.Computer simulations show that the proposed algorithm and activation function are effective on extracting features and improving the learning speed and recognition rate.