摘要
为了提高基于BP神经网络的手写体数字识别分类器的准确率与训练速度,针对基于BP神经网络的手写体数字识别分类器,从代价函数、权值初始化、正则方法消除过拟合几个方面对BP神经网络算法进行了改进。并使用MNIST数据集对分类器进行训练、验证、测试等实验。实验表明,改进后的手写体数字识别分类器的性能得到了优化。
In order to improve the accuracy and training speed of handwritten numeral recognition classifier based on BP neural network,for handwritten numeral recognition classifier based on BP neural network,We improve the algorithm from the following aspects, such as cost function,initialization of weights, regularization method. And we use MNIST data set classifier to train, validate, and test. Experiments show that the handwritten digit recognition classifier improved performance has been optimized.
出处
《电子设计工程》
2017年第6期27-30,共4页
Electronic Design Engineering
关键词
神经网络
BP算法
数字识别
分类器
neural network
BP algorithm
digit recognition
classifier
